No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Matrix factorization techniques for recommender systems
... This characteristic naturally encourages the development of iterative methods, which cycle over the alternated updates of U and V until convergence to a local stationary point. Similar strategies are commonly used in matrix completion and recommendation systems for the estimation of high-dimensional matrix factorization models; see, e.g., Zou et al. (2006), Koren et al. (2009) andMazumder et al. (2010). ...
... Despite being an effective and parallelizable strategy, in many applicative contexts, when it comes to factorizing huge matrices, an entire update of the parameters could be extremely expensive in terms of memory allocation and execution time. This is a well-understood issue in the literature on recommendation systems, where standard matrix completion problems may involve matrices with millions of rows and columns; see, e.g., Koren et al. (2009), Mairal et al. (2010, Recht and Ré (2013), Mensch et al. (2017). Moreover, batch optimization strategies do not generalize well to stream data contexts, where the data arrive sequentially and the parameters have to be updated on-the-fly as a new set of observations comes in. ...
... Then, having an error measure for a grid of ranks, we can pick the value minimizing the error. Such an approach is very popular in the machine-learning community for evaluating the performances of recommendation systems, collaborative filtering and matrix completion models, where the main goal is to obtain an accurate reconstruction of the unobserved entries of the response matrix; see, e.g., Koren et al. (2009) and Owen and Perry (2009) for an extensive statistically flavoured discussion. Rank selection based on error minimization is intuitive, general and fairly robust with respect to the optimization method used for the estimation. ...
Single-cell RNA sequencing allows the quantitation of gene expression at the individual cell level, enabling the study of cellular heterogeneity and gene expression dynamics. Dimensionality reduction is a common preprocessing step to simplify the visualization, clustering, and phenotypic characterization of samples. This step, often performed using principal component analysis or closely related methods, is challenging because of the size and complexity of the data. In this work, we present a generalized matrix factorization model assuming a general exponential dispersion family distribution and we show that many of the proposed approaches in the single-cell dimensionality reduction literature can be seen as special cases of this model. Furthermore, we propose a scalable adaptive stochastic gradient descent algorithm that allows us to estimate the model efficiently, enabling the analysis of millions of cells. Our contribution extends to introducing a novel warm start initialization method, designed to accelerate algorithm convergence and increase the precision of final estimates. Moreover, we discuss strategies for dealing with missing values and model selection. We benchmark the proposed algorithm through extensive numerical experiments against state-of-the-art methods and showcase its use in real-world biological applications. The proposed method systematically outperforms existing methods of both generalized and non-negative matrix factorization, demonstrating faster execution times while maintaining, or even enhancing, matrix reconstruction fidelity and accuracy in biological signal extraction. Finally, all the methods discussed here are implemented in an efficient open-source R package, sgdGMF, available at github/CristianCastiglione/sgdGMF
... In general, the user-item historical interactions are considered to be an important piece of information for expressing users' preferences, and then can be encoded to learn representations of users and items in recommendation systems [5], [6]. Over the past decade, with the development of online social networking services (e.g., Facebook, Weibo, Douban), recent studies show that social relations between users can be incorporated into enhancing the representation learning of users for recommendation systems [2], [8]- [11], which is motivated by the observation that users' preferences are likely to be influenced by their socially connected friends (e.g., classmates, colleagues, family members, etc.) [12]. ...
... Among all the techniques used in modern recommendation systems, collaborative filtering (CF) is one of the most widely used methods due to its outstanding performance in various recommendation scenarios. For example, MF [5] parameterized the ID of each user and item by a vectorized representation, and calculates user's preferences for items by conducting inner product between user and item. In practice, instead of expressing preferences directly in terms of ratings, users also express their preferences for items implicitly by clicking or browsing. ...
... To evaluate the effectiveness, we compare our proposed method with three groups of representative baselines, including recommendation systems without social network information (MF [5], NeuMF [1], NGCF [51], LightGCN [25]), social recommendation systems (SocialMF [8], LightGCN-S, DiffNet [10]), and self-supervised recommendation systems (SGL [21], MHCN [7]). These baseline methods are summarized as follows: ...
In recent years, researchers have attempted to exploit social relations to improve the performance in recommendation systems. Generally, most existing social recommendation methods heavily depends on substantial domain knowledge and expertise in primary recommendation tasks for designing useful auxiliary tasks. Meanwhile, Self-Supervised Learning (SSL) recently has received considerable attention in the field of recommendation, since it can provide self-supervision signals in assisting the improvement of target recommendation systems by constructing self-supervised auxiliary tasks from raw data without human-annotated labels. Despite the great success, these SSL-based social recommendations are insufficient to adaptively balance various self-supervised auxiliary tasks, since assigning equal weights on various auxiliary tasks can result in sub-optimal recommendation performance, where different self-supervised auxiliary tasks may contribute differently to improving the primary social recommendation across different datasets. To address this issue, in this work, we propose Adaptive Self-supervised Learning for Social Recommendations (AdasRec) by taking advantage of various self-supervised auxiliary tasks. More specifically, an adaptive weighting mechanism is proposed to learn adaptive weights for various self-supervised auxiliary tasks, so as to balance the contribution of such self-supervised auxiliary tasks for enhancing representation learning in social recommendations. The adaptive weighting mechanism is used to assign different weights on auxiliary tasks to achieve an overall weighting of the entire auxiliary tasks and ultimately assist the primary recommendation task, achieved by a meta learning optimization problem with an adaptive weighting network. Comprehensive experiments on various real-world datasets are constructed to verify the effectiveness of our proposed method.
... In light of these limitations, we propose a novel top-K recommendation method integrating multiple user feedback with popularity awareness. To start, we adopt the concept of the latent factor model, which learns to map users and items into corresponding lowrank feature spaces [4,24]. Then, these dense latent factors enable similarity computations between arbitrary pairs of users and items and thus can address the data sparsity prob- lem. ...
... These compact latent factors enable similarity computations between arbitrary user-item pairs, thereby addressing the data sparsity issue. As a representative example of LF models, matrix factorization (MF) methods model the preference of user u over item i as the inner product of their latent representations [24]. Various works can be seen as a variation of MF methods by employing different models to learn the user-item interaction function [5] or using different loss functions [4]. ...
... A well-established technique for estimating user preference is through latent factor models like matrix factorization [24], which is still among the most widely used and also the foundation of many state-of-the-art recommendation models [5,40,41]. Specifically, suppose the data we have is a binary user-item interaction matrix X ∈ {0, 1} m1×m2 encoding historical sales transactions, and the goal for now is to estimate the underlying purchase preference matrixX ∈ R m1×m2 . ...
Top-K recommendation involves inferring latent user preferences and generating personalized recommendations accordingly, which is now ubiquitous in various decision systems. Nonetheless, recommender systems usually suffer from severe \textit{popularity bias}, leading to the over-recommendation of popular items. Such a bias deviates from the central aim of reflecting user preference faithfully, compromising both customer satisfaction and retailer profits. Despite the prevalence, existing methods tackling popularity bias still have limitations due to the considerable accuracy-debias tradeoff and the sensitivity to extensive parameter selection, further exacerbated by the extreme sparsity in positive user-item interactions. In this paper, we present a \textbf{Pop}ularity-aware top-K recommendation algorithm integrating multi-behavior \textbf{S}ide \textbf{I}nformation (PopSI), aiming to enhance recommendation accuracy and debias performance simultaneously. Specifically, by leveraging multiple user feedback that mirrors similar user preferences and formulating it as a three-dimensional tensor, PopSI can utilize all slices to capture the desiring user preferences effectively. Subsequently, we introduced a novel orthogonality constraint to refine the estimated item feature space, enforcing it to be invariant to item popularity features thereby addressing our model's sensitivity to popularity bias. Comprehensive experiments on real-world e-commerce datasets demonstrate the general improvements of PopSI over state-of-the-art debias methods with a marginal accuracy-debias tradeoff and scalability to practical applications. The source code for our algorithm and experiments is available at \url{https://github.com/Eason-sys/PopSI}.
... We evaluate the effectiveness of our RecLM approach by integrating it with state-of-the-art recommender systems, allowing us to assess performance improvements in a model-agnostic manner compared to baseline models. The selected CF recommenders include non-graph methods such as BiasMF (Koren et al., 2009) and NCF (He et al., 2017), the GNN-enhanced method LightGCN (He et al., 2020), and graph contrastive learning approaches SGL (Wu et al., 2021) and SimGCL (Yu et al., 2022). Details regarding the datasets and baseline methods are provided in Appendices 6.1 and 6.2. ...
... In recommender systems, numerous collaborative filtering models have been proposed to map users and items into latent representations based on user/item IDs (Koren et al., 2021;Su & Khoshgoftaar, 2009). These methods have evolved significantly, starting from early matrix factorization techniques, such as BiasMF (Koren et al., 2009), to the introduction of Neural Collaborative Filtering (NCF) with the advent of neural networks (He et al., 2017). Recently, advancements in Graph Neural Networks (GNNs) have opened promising avenues for constructing bipartite graphs based on useritem interaction history, allowing for the capture of high-order collaborative relationships. ...
... • BiasMF (Koren et al., 2009): A matrix factorization-based approach that enhances recommendation accuracy by incorporating user and item bias vectors to capture individual preference patterns. ...
Modern recommender systems aim to deeply understand users' complex preferences through their past interactions. While deep collaborative filtering approaches using Graph Neural Networks (GNNs) excel at capturing user-item relationships, their effectiveness is limited when handling sparse data or zero-shot scenarios, primarily due to constraints in ID-based embedding functions. To address these challenges, we propose a model-agnostic recommendation instruction-tuning paradigm that seamlessly integrates large language models with collaborative filtering. Our proposed Recommendation Language Model (RecLM) enhances the capture of user preference diversity through a carefully designed reinforcement learning reward function that facilitates self-augmentation of language models. Comprehensive evaluations demonstrate significant advantages of our approach across various settings, and its plug-and-play compatibility with state-of-the-art recommender systems results in notable performance enhancements.
... Learning low rank matrices is a problem of broad interest in machine learning and statistics, with applications ranging from collaborative filtering [1,2], to multitask learning [3], to computer vision [4], and many more. A principled approach to tackle this problem is via suitable convex relaxations. ...
... 1 University College London, London, UK 2 Computational Statistics and Machine Learning -Istituto Italiano di Tecnologia, Genova, Italy Due to the above shortcomings, practical algorithms for low rank matrix completion often use an explicit low rank matrix factorization to reduce the number of variables (see e.g. [2,9] and references therein). In particular, a reduced variational form of the trace norm is used [7]. ...
... However, at each iteration PFB performs the SVD of an n × m matrix via Eqs. (2) and (3), which requires O(min(n, m) nm) operations, a procedure that becomes prohibitively expensive for large values of n and m. Other methods such as those based on Frank-Wolfe procedure (e.g. ...
Trace norm regularization is a widely used approach for learning low rank matrices. A standard optimization strategy is based on formulating the problem as one of low rank matrix factorization which, however, leads to a non-convex problem. In practice this approach works well, and it is often computationally faster than standard convex solvers such as proximal gradient methods. Nevertheless, it is not guaranteed to converge to a global optimum, and the optimization can be trapped at poor stationary points. In this paper we show that it is possible to characterize all critical points of the non-convex problem. This allows us to provide an efficient criterion to determine whether a critical point is also a global minimizer. Our analysis suggests an iterative meta-algorithm that dynamically expands the parameter space and allows the optimization to escape any non-global critical point, thereby converging to a global minimizer. The algorithm can be applied to problems such as matrix completion or multitask learning, and our analysis holds for any random initialization of the factor matrices. Finally, we confirm the good performance of the algorithm on synthetic and real datasets.
... In such cases, it becomes challenging to provide accurate recommendations as the system lacks the necessary user preferences or item characteristics. Techniques like content-based filtering or hybrid approaches are often used to address this problem and make initial recommendations based on available data or user attributes [13,14]. ...
... This leads to the problem of finding reliable similarity measures which can affect the accuracy of collaborative filtering techniques. Matrix factorization-based methods, such as singular value decomposition (SVD) or matrix completion techniques can overcome this limitation [13,15]. ...
... Recommender systems (RS) are increasingly important in social media, entertainment, and e-commerce in the information explosion era (Shi, Larson, and Hanjalic 2014;Wang et al. 2020a). It provides personalized recommendations to users by analyzing their historical behavior (Koren, Bell, and Volinsky 2009;He et al. 2017;Zhang, Liu, and Wu 2018;Wang et al. 2021a). However, the collected data contains many biases such as selection bias, as users are free to choose items to rate, making the collected data not representative of the target population (Marlin and Zemel 2009;Marlin et al. 2007;Wang et al. 2024b;Yang et al. 2023), which challenges the unbiased learning of RS. ...
... We compared our method with a series of baseline methods widely utilized in debiasing (RS), including matrix factorization (MF) (Koren, Bell, and Volinsky 2009) ...
Recommender systems (RS) are increasingly important in social media, entertainment, and e-commerce in the information explosion era. However, the collected data contains many biases such as selection bias, as users are free to choose items to rate, making the collected data not representative of the target population. Recently, many methods such as relabeling-based and reweighting-based have been proposed to mitigate the selection bias. However, the effectiveness of these methods relies on strong assumptions, which are difficult to satisfy in real-world scenarios, leading to sub-optimal debiasing performance. In this paper, we propose a debiasing method from the machine unlearning perspective. Specifically, we first propose a user unlearning rate network to determine which user needs to be unlearned. Then we generate the error-maximizing pseudo-labels for each user and fusion such pseudo-labels and the observed labels based on the learned user unlearning rate to mitigate the selection bias. In addition, we further propose an unlearning to debias training algorithm to achieve unbiased learning of the prediction model. Finally, we conduct extensive experiments on three real-world datasets to validate the effectiveness of our method.
... • MF (Koren, Bell, and Volinsky, 2009): Matrix Factorization is a widely used model in the recommendation, which is a biased model without debiasing capability. ...
... We conducted 20 repeated experiments with different random seeds in the same settings to evaluate the stability of models. We compute the standard deviation of the performance on each model on Yahoo!R3 dataset, and choose the base model (i.e., MF (Koren et al., 2009) ) in these methods as the baseline to compare with. As shown in Fig. 3, there is a certain gap between InvPref and the base model MF, while the standard deviations of the model performance in InterD and AutoDebias are 13 times larger than MF, showing the instability of these methods. ...
Previous debiasing studies utilize unbiased data to make supervision of model training. They suffer from the high trial risks and experimental costs to obtain unbiased data. Recent research attempts to use invariant learning to detach the invariant preference of users for unbiased recommendations in an unsupervised way. However, it faces the drawbacks of low model accuracy and unstable prediction performance due to the losing cooperation with variant preference. In this paper, we experimentally demonstrate that invariant learning causes information loss by directly discarding the variant information, which reduces the generalization ability and results in the degradation of model performance in unbiased recommendations. Based on this consideration, we propose a novel lightweight knowledge distillation framework (KDDebias) to automatically learn the unbiased preference of users from both invariant and variant information. Specifically, the variant information is imputed to the invariant user preference in the distance-aware knowledge distillation process. Extensive experiments on three public datasets, i.e., Yahoo!R3, Coat, and MIND, show that with the biased imputation from the variant preference of users, our proposed method achieves significant improvements with less than 50% learning parameters compared to the SOTA unsupervised debiasing model in recommender systems. Our code are publicly available at https://github.com/BAI-LAB/KD-Debias.
... Link prediction can predict the presence of a relationship (edge) between two individuals (nodes) in a social network, either presently or in the near future [19]. Recommendation systems try to recommend products to customers; this task is a link prediction problem, where one seeks edges between two different types of nodes, the product nodes and the customer nodes [20,21]. Link prediction for entity resolution predicts links between different records in a dataset that refer to the same object [22,23]. ...
... Sometimes called the update function, ϕ : Rl k → Rl k+1 is a differentiable function with trainable parameters. A common choice for ϕ is a single layer, feedforward neural network, as in (20). The term is a permutation-invariant aggregation function such as an element-wise vector-valued sum, mean or maximum, N i is the 1-hop neighborhood of i (excluding i), and the term µ ij ∈ Rl k is a feature vector (defined below) that describes the interaction of node i with node j. ...
Graph neural networks are deep neural networks designed for graphs with attributes attached to nodes or edges. The number of research papers in the literature concerning these models is growing rapidly due to their impressive performance on a broad range of tasks. This survey introduces graph neural networks through the encoder-decoder framework and provides examples of decoders for a range of graph analytic tasks. It uses theory and numerous experiments on homogeneous graphs to illustrate the behavior of graph neural networks for different training sizes and degrees of graph complexity.
... At the core of modern collaborative recommendation is learning high-quality representations of users and items from user-item interaction data. Matrix factorization [21,35] offered a first step towards representation learning for recommendation, projecting user and item IDs into an embedding space and performing predictions based on these vectors. Subsequent work built on these foundations, augmenting the single user or item ID by incorporating additional information into the representations, such as interaction history [16,20] and rich side information [34,36]. ...
... 3 . 3 is used in Eq. (21) to control the ratio of mixing between local and long-range/global collaborative information. We give the tuning results of 3 on LastFM and Amazon-book datasets in Figure 8. ...
Collaborative recommendation fundamentally involves learning high-quality user and item representations from interaction data. Recently, graph convolution networks (GCNs) have advanced the field by utilizing high-order connectivity patterns in interaction graphs, as evidenced by state-of-the-art methods like PinSage and LightGCN. However, one key limitation has not been well addressed in existing solutions: capturing long-range collaborative filtering signals, which are crucial for modeling user preference. In this work, we propose a new graph transformer (GT) framework -- \textit{Position-aware Graph Transformer for Recommendation} (PGTR), which combines the global modeling capability of Transformer blocks with the local neighborhood feature extraction of GCNs. The key insight is to explicitly incorporate node position and structure information from the user-item interaction graph into GT architecture via several purpose-designed positional encodings. The long-range collaborative signals from the Transformer block are then combined linearly with the local neighborhood features from the GCN backbone to enhance node embeddings for final recommendations. Empirical studies demonstrate the effectiveness of the proposed PGTR method when implemented on various GCN-based backbones across four real-world datasets, and the robustness against interaction sparsity as well as noise.
... Traditional cognitive diagnosis. A significant body of literature has been devoted to cognitive diagnosis, including deterministic input, noise, and gate models (DINA) [15], item response theory (IRT) [16], multidimensional IRT (MIRT) [17], and matrix factorization (MF) [18]. Despite demonstrating some effectiveness, these approaches rely on manually defined interaction functions, which combine features of learners and exercises through multiplicative terms, such as logical functions [16] or inner products [18]. ...
... A significant body of literature has been devoted to cognitive diagnosis, including deterministic input, noise, and gate models (DINA) [15], item response theory (IRT) [16], multidimensional IRT (MIRT) [17], and matrix factorization (MF) [18]. Despite demonstrating some effectiveness, these approaches rely on manually defined interaction functions, which combine features of learners and exercises through multiplicative terms, such as logical functions [16] or inner products [18]. However, this may be insufficient to capture the complex relationships between learners and exercises [19]. ...
Cognitive diagnosis represents a fundamental research area within intelligent education, with the objective of measuring the cognitive status of individuals. Theoretically, an individual's cognitive state is essentially equivalent to their cognitive structure state. Cognitive structure state comprises two key components: knowledge state (KS) and knowledge structure state (KUS). The knowledge state reflects the learner's mastery of individual concepts, a widely studied focus within cognitive diagnosis. In contrast, the knowledge structure state-representing the learner's understanding of the relationships between concepts-remains inadequately modeled. A learner's cognitive structure is essential for promoting meaningful learning and shaping academic performance. Although various methods have been proposed, most focus on assessing KS and fail to assess KUS. To bridge this gap, we propose an innovative and effective framework-CSCD (Cognitive Structure State-based Cognitive Diagnosis)-which introduces a novel framework to modeling learners' cognitive structures in diagnostic assessments, thereby offering new insights into cognitive structure modeling. Specifically, we employ an edge-feature-based graph attention network to represent the learner's cognitive structure state, effectively integrating KS and KUS. Extensive experiments conducted on real datasets demonstrate the superior performance of this framework in terms of diagnostic accuracy and interpretability.
... Broadly, existing graph learning methods can be categorized by their use of graphs during preprocessing, training, or inference [58]. In RecSys, traditional collaborative filtering and GNN-based methods use the graph during training [27,34,49], while recent methods like TAG-CF leverage it during test-time inference. [30]. ...
... Backbones 2 . We use matrix factorization (MF) [34], Neural Matrix Factorization (NeuMF) [28], LightGCN [27], and MF+DirectAU (DAU) loss [47] as backbones for the retrieval task, where the first three are trained with the Bayesian Personalized Ranking (BPR) loss [43]; we use WideDeep [9], DLRM [39], and DCNv2 [48], all trained with binary cross entropy loss (LogLoss) for the CTR task. ...
Deep recommender systems rely heavily on large embedding tables to handle high-cardinality categorical features such as user/item identifiers, and face significant memory constraints at scale. To tackle this challenge, hashing techniques are often employed to map multiple entities to the same embedding and thus reduce the size of the embedding tables. Concurrently, graph-based collaborative signals have emerged as powerful tools in recommender systems, yet their potential for optimizing embedding table reduction remains unexplored. This paper introduces GraphHash, the first graph-based approach that leverages modularity-based bipartite graph clustering on user-item interaction graphs to reduce embedding table sizes. We demonstrate that the modularity objective has a theoretical connection to message-passing, which provides a foundation for our method. By employing fast clustering algorithms, GraphHash serves as a computationally efficient proxy for message-passing during preprocessing and a plug-and-play graph-based alternative to traditional ID hashing. Extensive experiments show that GraphHash substantially outperforms diverse hashing baselines on both retrieval and click-through-rate prediction tasks. In particular, GraphHash achieves on average a 101.52% improvement in recall when reducing the embedding table size by more than 75%, highlighting the value of graph-based collaborative information for model reduction.
... The collaborative filtering (CF) is the most adapted technique in RS. Traditional CF models such as matrix factorization (MF) [25], probabilistic MF (PMF) [39], singular value decomposition (SVD) [51], etc., use rating data to predict user-interested items. Similarly, in [39], proposed a probabilistic approach for the user interest prediction. ...
... The MF, SVD, and PMF approaches are the traditional existing approaches that work based on user ratings. The SBMF+R [52] approach is to predict [ 25] 1) This approach is a standard factorization approach, and 2) We would like to compare this approach as a basic latent factor-based model 2 SVD [51] 1) It is a standard decomposition technique predicted using latent factors, and 2) To compare the accuracy of SVD with user rating data and consider both user ratings and user reviews 3 P M F [ 39] 1) It is a classical probabilistic approach and predicted interests using latent factors, and 2) To compare as a basic probabilistic model ...
Recommender system (RS) predicts relevant items to a new user based on the purchase history of existing users. In this area, Collaborative Filtering (CF) is one of the popular approaches, uses the user rating data for recommendations. Generally, users express their opinion on the items purchased through user ratings. Considering the user rating data alone may not be sufficient to identify the user interests that are nearer to the user intent. So, there is a need to look into other data provided in the purchase history such as user reviews, user profile, and helpfulness. In this paper, we propose an approach Elaboration Likelihood Model for Discrepant-users (ELM-D) to address the issue of identifying user interest using user rating data and user review data. In this approach, we use elaboration to identify the correct data to reach the user interests nearer to the user intent for discrepant-users. We observed real-world datasets and found the discrepant-users. The discrepant-users are the users who do not provide user ratings and/or reviews correctly. To understand the user interests from user review data, we proposed an approach to extract the user emotions using emotion detection algorithm by exploiting n-polarity. We built a CF approach to predict the interest of a new user using user rating data and user emotions from user review data. We conducted experiments on the real-world datasets from Amazon and Yelp. We evaluated results using mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) metrics. The proposed approach outperforms the existing approaches.
... Biases in recommendation algorithms are rooted in the design and limitations of widely adopted techniques, such as matrix factorization (MF) and deep learning-based models (Zhang et al., 2019). These approaches excel at capturing complex user-item interactions but often fail to address fairness, as their underlying latent factor models emphasize patterns driven by popularity (Koren et al., 2009). Prototype-based recommender systems, such as Prototype-based Matrix Factorization (ProtoMF) (Melchiorre et al., 2022), were introduced to enhance explainability by identifying prototypes representing typical behaviors. ...
... To evaluate our model's performance, we select four baselines: Matrix Factorization (MF) (Koren et al., 2009), anchor-based collaborative filtering (ACF) (Barkan et al., 2021), ProtoMF (Melchiorre et al., 2022). Additionally, we include a benchmark for popularity bias mitigation, ZeroSum (Rhee et al., 2022), by adding the score difference regularizer to the loss function of user and item side of ProtoMF. ...
Popularity bias in recommender systems can increase cultural overrepresentation by favoring norms from dominant cultures and marginalizing underrepresented groups. This issue is critical for platforms offering cultural products, as they influence consumption patterns and human perceptions. In this work, we address popularity bias by identifying demographic biases within prototype-based matrix factorization methods. Using the country of origin as a proxy for cultural identity, we link this demographic attribute to popularity bias by refining the embedding space learning process. First, we propose filtering out irrelevant prototypes to improve representativity. Second, we introduce a regularization technique to enforce a uniform distribution of prototypes within the embedding space. Across four datasets, our results demonstrate a 27\% reduction in the average rank of long-tail items and a 2\% reduction in the average rank of items from underrepresented countries. Additionally, our model achieves a 2\% improvement in HitRatio@10 compared to the state-of-the-art, highlighting that fairness is enhanced without compromising recommendation quality. Moreover, the distribution of prototypes leads to more inclusive explanations by better aligning items with diverse prototypes.
... With the explosive growth of Internet information, recommendation systems have been playing an increasingly important role in on-line E-commerce and applications in a variety of areas, including music streaming service such as Spotify 1 and Apple Music, movie rating such as IMDB 2 , video streaming service such as Netflix and Youtube, job recommendation such as LinkedIn 3 , and product recommendation such as Amazon. Many recommendation methods are based on Collaborative Filtering (CF) which mainly makes use of historical ratings [14,15,18,22,31,33,35]. Recently, some approaches also consider text information in addition to the rating data [1,21,23,26,40,49]. ...
... Collaborative filtering (CF) has been studied for a long time and has achieved some success in recommendation systems [27,37]. Latent Factor Models (LFM) based on Matrix Factorization (MF) [15] play an important role for rating prediction. Various MF algorithms have been proposed, such as Singular Value Decomposition (SVD) and SVD++ [14], Non-negative Matrix Factorization (NMF) [18], and Probabilistic Matrix Factorization (PMF) [31]. ...
Recently, some E-commerce sites launch a new interaction box called Tips on their mobile apps. Users can express their experience and feelings or provide suggestions using short texts typically several words or one sentence. In essence, writing some tips and giving a numerical rating are two facets of a user's product assessment action, expressing the user experience and feelings. Jointly modeling these two facets is helpful for designing a better recommendation system. While some existing models integrate text information such as item specifications or user reviews into user and item latent factors for improving the rating prediction, no existing works consider tips for improving recommendation quality. We propose a deep learning based framework named NRT which can simultaneously predict precise ratings and generate abstractive tips with good linguistic quality simulating user experience and feelings. For abstractive tips generation, gated recurrent neural networks are employed to "translate" user and item latent representations into a concise sentence. Extensive experiments on benchmark datasets from different domains show that NRT achieves significant improvements over the state-of-the-art methods. Moreover, the generated tips can vividly predict the user experience and feelings.
... Matrix Completion (MC) methods have found widespread application in the estimation of traffic data. Some algorithms, such as the convex relaxation method based on minimum nuclear norm approximation [12] and matrix factorization-based methods [13], leverage the linear spatiotemporal characteristics of traffic data to infer missing values. However, these methods are often too simplistic, which can lead to inaccurate estimations when applied to large-scale traffic data. ...
The operations and maintenance of satellite networks heavily depend on traffic measurements. Due to the large-scale and highly dynamic nature of satellite networks, global measurement encounters significant challenges in terms of complexity and overhead. Estimating global network traffic data from partial traffic measurements is a promising solution. However, the majority of current estimation methods concentrate on low-rank linear decomposition, which is unable to accurately estimate. The reason lies in its inability to capture the intricate nonlinear spatio-temporal relationship found in large-scale, highly dynamic traffic data. This paper proposes Satformer, an accurate and robust method for estimating traffic data in satellite networks. In Satformer, we innovatively incorporate an adaptive sparse spatio-temporal attention mechanism. In the mechanism, more attention is paid to specific local regions of the input tensor to improve the model's sensitivity on details and patterns. This method enhances its capability to capture nonlinear spatio-temporal relationships. Experiments on small, medium, and large-scale satellite networks datasets demonstrate that Satformer outperforms mathematical and neural baseline methods notably. It provides substantial improvements in reducing errors and maintaining robustness, especially for larger networks. The approach shows promise for deployment in actual systems.
... However, those methods are unable to capture complex patterns or correlations between multiple variables. Due to those limitations, other imputation methods were developed to impute missing data with greater accuracy, such as kernel methods, matrix completion and factorization [KBV09,MHT10,YRD16]. ...
Missing data in time-series analysis poses significant challenges, affecting the reliability of downstream applications. Imputation, the process of estimating missing values, has emerged as a key solution. This paper introduces BRATI, a novel deep-learning model designed to address multivariate time-series imputation by combining Bidirectional Recurrent Networks and Attention mechanisms. BRATI processes temporal dependencies and feature correlations across long and short time horizons, utilizing two imputation blocks that operate in opposite temporal directions. Each block integrates recurrent layers and attention mechanisms to effectively resolve long-term dependencies. We evaluate BRATI on three real-world datasets under diverse missing-data scenarios: randomly missing values, fixed-length missing sequences, and variable-length missing sequences. Our findings demonstrate that BRATI consistently outperforms state-of-the-art models, delivering superior accuracy and robustness in imputing multivariate time-series data.
... These methods map various object features to a low-dimensional space through matrix factorization, extracting latent features from complex data to predict interactions between different entities. Matrix factorization excels at extracting latent features from intricate data, revealing the structural relationships within the data [18]. However, when facing high-dimensional data or lower data quality, matrix-based methods may encounter performance issues [19]. ...
The process of new drug development is complex, whereas drug-disease association (DDA) prediction aims to identify new therapeutic uses for existing medications. However, existing graph contrastive learning approaches typically rely on single-view contrastive learning, which struggle to fully capture drug-disease relationships. Subsequently, we introduce a novel multi-view contrastive learning framework, named CDPMF-DDA, which enhances the model's ability to capture drug-disease associations by incorporating diverse information representations from different views. First, we decompose the original drug-disease association matrix into drug and disease feature matrices, which are then used to reconstruct the drug-disease association network, as well as the drug-drug and disease-disease similarity networks. This process effectively reduces noise in the data, establishing a reliable foundation for the networks produced. Next, we generate multiple contrastive views from both the original and generated networks. These views effectively capture hidden feature associations, significantly enhancing the model's ability to represent complex relationships. Extensive cross-validation experiments on three standard datasets show that CDPMF-DDA achieves an average AUC of 0.9475 and an AUPR of 0.5009, outperforming existing models. Additionally, case studies on Alzheimer’s disease and epilepsy further validate the model’s effectiveness, demonstrating its high accuracy and robustness in drug-disease association prediction. Based on a multi-view contrastive learning framework, CDPMF-DDA is capable of integrating multi-source information and effectively capturing complex drug-disease associations, making it a powerful tool for drug repositioning and the discovery of new therapeutic strategies.
... We compared CDLD with NCF [36] and MF [47] methods. By comparing CDLD with both MF and NCF, we aim to demonstrate how CDLD not only addresses the limitations of traditional linear models like MF but also enhances deep learning techniques exemplified by NCF through its unique cyclic training mechanism. ...
Recommendation systems are tasked with the complex challenge of modeling high-dimensional interactions between users and items to deliver personalized recommendations. This paper introduces Cyclic Dual Latent Discovery (CDLD), a novel method that employs dual deep neural networks (DNNs) in a cyclic training process to discover the latent traits of users and items based on their interactions. CDLD operates on the principle that interactions between users and items inherently possess information about their latent traits. By cyclically refining these traits through two interconnected DNNs, CDLD effectively captures the non-linear relationships between them. When evaluated on the MovieLens 100K dataset, CDLD demonstrated superior performance, achieving a root mean square error (RMSE) that is 2.86% lower than that of matrix factorization and 2.51% lower than that of neural collaborative filtering, thus showing strong generalization capabilities. Additionally, CDLD addresses the cold-start problem effectively by adjusting only the new entity’s latent traits during training, while maintaining fixed model weights, even with limited data samples. The proposed method not only enhances the accuracy of recommendations but also improves the scalability and robustness of the model. Through its cyclical training mechanism, CDLD ensures ongoing refinement of latent traits, paving the way for more adaptive and efficient recommendation systems across diverse applications.
... (i) MF[27]: MF is a classic user-item recommendation framework based on matrix factorization. (ii) DeepFM[16]: DeepFM adds extra neural networks based on MF and recommends items by considering the combination of the scores of neural networks and MF. ...
Current recommendation systems recommend goods by considering users' historical behaviors, social relations, ratings, and other multi-modals. Although outdated user information presents the trends of a user's interests, no recommendation system can know the users' real-time thoughts indeed. With the development of brain-computer interfaces, it is time to explore next-generation recommenders that show users' real-time thoughts without delay. Electroencephalography (EEG) is a promising method of collecting brain signals because of its convenience and mobility. Currently, there is only few research on EEG-based recommendations due to the complexity of learning human brain activity. To explore the utility of EEG-based recommendation, we propose a novel neural network model, QUARK, combining Quantum Cognition Theory and Graph Convolutional Networks for accurate item recommendations. Compared with the state-of-the-art recommendation models, the superiority of QUARK is confirmed via extensive experiments.
... The evolution of collaborative filtering has spawned diverse approaches, from classical matrix factorization methods (e.g. [13]) to sophisticated neural architectures (e.g. [9]). ...
Graph neural networks (GNNs) have demonstrated superior performance in collaborative recommendation through their ability to conduct high-order representation smoothing, effectively capturing structural information within users' interaction patterns. However, existing GNN paradigms face significant challenges in scalability and robustness when handling large-scale, noisy, and real-world datasets. To address these challenges, we present LightGNN, a lightweight and distillation-based GNN pruning framework designed to substantially reduce model complexity while preserving essential collaboration modeling capabilities. Our LightGNN framework introduces a computationally efficient pruning module that adaptively identifies and removes redundant edges and embedding entries for model compression. The framework is guided by a resource-friendly hierarchical knowledge distillation objective, whose intermediate layer augments the observed graph to maintain performance, particularly in high-rate compression scenarios. Extensive experiments on public datasets demonstrate LightGNN's effectiveness, significantly improving both computational efficiency and recommendation accuracy. Notably, LightGNN achieves an 80% reduction in edge count and 90% reduction in embedding entries while maintaining performance comparable to more complex state-of-the-art baselines. The implementation of our LightGNN framework is available at the github repository: https://github.com/HKUDS/LightGNN.
... Regarding the temporal consideration, some algorithms [32] use explicit timestamps to build models that understand past behavior based on a specific time of a user's preference. Some algorithms do not directly use specific timestamps, but model the sequence of actions directly. ...
Sequential recommendation refers to recommending the next item of interest for a specific user based on his/her historical behavior sequence up to a certain time. While previous research has extensively examined Markov chain-based sequential recommendation models, the majority of these studies has focused on the user's historical behavior sequence but has paid little attention to the overall correlation between items. This study introduces a sequential recommendation algorithm known as Item Association Factorization Mixed Markov Chains, which incorporates association information between items using an item association graph, integrating it with user behavior sequence information. Our experimental findings from the four public datasets demonstrate that the newly introduced algorithm significantly enhances the recommendation ranking results without substantially increasing the parameter count. Additionally, research on tuning the prior balancing parameters underscores the significance of incorporating item association information across different datasets.
... Among various collaborative filtering recommendation algorithms, matrix factorization is one of the popular ones. Matrix factorization decomposes user and item interaction information into two low dimensional user matrix and item matrix, obtains the potential relationship between users and items, and recommends items of interest to users through the product of the user matrix and item matrix [5]. In recent years, deep learning has also developed rapidly, and neural networks in deep learning have achieved great success in many fields. ...
The neural collaborative filtering recommendation algorithm is widely used in recommendation systems, which further applies deep learning to recommendation systems. It is a universal framework in the neural collaborative filtering recommendation algorithm, however, it does not consider the impact of different features on recommendation results, nor does it consider the issues of data sparsity and long tail distribution of items. To solve the above problems, this paper proposes a recommendation algorithm based on attention mechanism and contrastive learning, which focuses on more important features through attention mechanism and increases the number of samples to achieve data augmentation through contrastive learning, therefore it improves model performance. The experimental results on two benchmark datasets show that the algorithm proposed in this paper has further improved recommendation performance compared to other benchmark algorithms.
... Dado que cada usuario ha calificado algunas películas, nos gustaría predecir la calificación que un usuario podría darle a un elemento que aún no ha sido calificado. Según las calificaciones predichas por la factorización matricial, se dan recomendaciones al usuario (Chen and Wang 2022;Koren et al. 2009). Vol. ...
The rise of electronic payments and their development through national gateways represent a challenge for decision-making aimed at promoting payments for services. This rapid expansion has led to increasing use of machine learning models that allow analyzing customer characteristics and enhancing the services provided. A solution to this responds to recommendation systems which can be of great benefit for applications such as Transfermóvil. This article demonstrates the use of these systems through the matrix factorization algorithm and how it can be used in this payment gateway. Furthermore, to apply its use, an AutoML tool is used that facilitates the use of the algorithm.
... These systems employ a combination of techniques, including collaborative filtering, content-based filtering, and hybrid methods, to analyze user data and generate personalized recommendation [5]. Collaborative filtering is widely utilized by platforms like Netflix and Amazon to offer suggestions by examining the behaviour patterns of similar users such as their ratings and purchase history [10]. Nevertheless, the substantial data gathering necessary for the operation of such systems gives rise to notable apprehensions regarding privacy. ...
This study focuses on the influence of information exposure under an intelligent recommendation system on teenagers’ privacy protection behaviour. By employing the Communication Privacy Management (CPM) theory, we examine how information exposure through intelligent recommendation systems influences teenagers’ attitudes and actions towards privacy. The results of this study emphasize the need for the CPM advocated by Petronio [1] and the role of various factors in influencing privacy protection behaviour. The study also explores teenagers’ privacy protection behaviours on social media. Through quantitative research, this study provides a perspective on the interaction among adolescents’ digital literacy, privacy issues and behavioural responses in the context of intelligent recommendation systems.
... The similarity (e.g., dot product) between user and item representations is used to infer the predicted scores of unobserved user-item pairs. Conventional methods primarily rely on matrix factorization (MF) [11] and autoencoders [12]. Moreover, in recent years, graph neural networks (GNNs) have gained increasing interest, as they help capture higher-order relational information by treating observed user-item interactions as a bipartite graph. ...
In recommender systems, Graph Collaborative Filtering (GCF) is widely used for its ability to effectively model the interaction between users and items. However, in practical scenarios, GCF faces a significant problem: the representation of popular items tends to be over-concentrated, while cold items are marginalized, leading to recommendation results biased towards popular items and making it difficult to address the issue of long-tail distribution. To alleviate data sparsity, existing GCF methods typically incorporate Contrastive Learning (CL) to assist in updating node representations. However, inappropriate CL methods can introduce extra noise. For this reason, this paper proposes an Enhanced Contrastive Learning-based Graph Collaborative Filtering (ECL-GCF). The model improves the traditional GCF approach by: 1. capturing explicit interaction information between users and items by exploiting structural neighborhood contrastive learning; 2. introducing semantic neighborhood contrastive learning to construct potential similarity relationships by capturing implicit semantic information of users and items, thereby providing more meaningful representations for cold items; and 3. optimizing the embedding representation by regularizing chi-square and homogeneous embedding representations, ensuring that the embeddings are both close to positive sample pairs and uniformly distributed in the space, thus preventing the marginalization of cold items. Experimental results indicate that the model improves recommendation performance by approximately 5% on the Yelp2018 and iFashion datasets, and especially performs well on cold item recommendation.
... Among these systems, sequential recommendation (SR) methods (Wang et al., 2019;Zhou et al., 2018;Kang and McAuley, 2018) have gained prominence due to their ability to capture dynamic user interests more effectively than traditional collaborative filtering techniques. Mainstream SR approaches primarily rely on ID-based deep learning strategies (Koren et al., 2009;Goldberg et al., 1992), such as matrix factorization and deep sequence neural networks. These methods encode users and items as unique identifiers, leveraging historical interaction data to learn sequential behavioral patterns. ...
Sequential recommendation (SR) systems have evolved significantly over the past decade, transitioning from traditional collaborative filtering to deep learning approaches and, more recently, to large language models (LLMs). While the adoption of LLMs has driven substantial advancements, these models inherently lack collaborative filtering information, relying primarily on textual content data neglecting other modalities and thus failing to achieve optimal recommendation performance. To address this limitation, we propose Molar, a Multimodal large language sequential recommendation framework that integrates multiple content modalities with ID information to capture collaborative signals effectively. Molar employs an MLLM to generate unified item representations from both textual and non-textual data, facilitating comprehensive multimodal modeling and enriching item embeddings. Additionally, it incorporates collaborative filtering signals through a post-alignment mechanism, which aligns user representations from content-based and ID-based models, ensuring precise personalization and robust performance. By seamlessly combining multimodal content with collaborative filtering insights, Molar captures both user interests and contextual semantics, leading to superior recommendation accuracy. Extensive experiments validate that Molar significantly outperforms traditional and LLM-based baselines, highlighting its strength in utilizing multimodal data and collaborative signals for sequential recommendation tasks. The source code is available at https://anonymous.4open.science/r/Molar-8B06/.
... The CF methods mine the user behavioral similarities to generate recommendations by using the user-item rating matrix. For example, Koren et al. [10] decomposed the feature vectors of users and items from the rating matrix for prediction purpose. However, the CF methods also encounter challenges such as data sparsity and cold-start problems, which may make it difficult to calculate similarity due to insufficient interaction data [11]. ...
Knowledge graph (KG) with enriched items’ related information has been widely used to alleviate the data sparsity and cold-start problems in recommender systems. However, the noise in KG that is irrelevant to a recommendation task may mislead the decision outcomes. The existing research predominantly employs data-driven modeling which uncovers the underlying patterns of the model by mining correlations within the data. This learning paradigm that lacks causality may lead to spurious associations and limit the robustness of recommendation results. To tackle this problem, this paper proposes a novel framework called knowledge graph denoising-based causal recommendation (KGDCR). In this framework, we fully combine the advantages of data-driven and model-driven modeling, and introduce a causality-driven recommendation mechanism based on causal inference. This mechanism enhances the robustness of the model by identifying the causal relationships between user behaviors and recommendation decisions. Specifically, we leverage graph attention neural networks to aggregate semantic information from the KG. Furthermore, the KGDCR captures personalized user preferences at a fine granularity by intervening in the noise. Then, we formulate a cross-view-constrained optimization problem to guide the recommendation model towards stable prediction. Experimental results demonstrate that the denoising performance and robustness of the KGDCR outperform the existing methods.
... Traditional recommendation approaches, such as content-based filtering and collaborative filtering, have shown promise in various domains, including e-commerce and entertainment (Linden et al., 2003;Koren et al., 2009). However, the unique complexities of educational contexts-such as the long-term nature of learning goals, the interdependence of course topics, and the diversity of learning styles-call for more advanced techniques (Jannach et al., 2010;Bobadilla et al., 2013). ...
The proliferation of e-learning platforms has created a need for sophisticated course recommendation systems. This paper presents an innovative online course recommendation system using Neural Collaborative Filtering (NCF), a deep learning technique designed to surpass traditional methods in accuracy and personalization. Our system employs a hybrid NCF architecture, integrating matrix factorization with multi-layer perceptron to capture complex user-course interactions. The proposed NCF-based recommendation system aims to address key challenges in the e-learning domain, such as diverse user preferences, varying course content, and evolving learning patterns. By leveraging the power of neural networks, our approach seeks to provide more relevant and personalized course suggestions to learners. Our research contributes to the intersection of deep learning and educational technology, offering new insights into how advanced machine learning techniques can be applied to improve online learning experiences. The proposed system has the potential to enhance the quality of course recommendations, leading to more effective learning pathways for users. This work has important implications for e-learning platforms, educational institutions, and lifelong learners navigating the vast landscape of online courses. By improving the match between learners and courses, we aim to increase engagement, completion rates, and overall satisfaction in online education. Future work will explore the long-term impact of such personalized recommendations on learning outcomes and skill development.
... Their fundamental task is to predict a user's preference for an item. Traditional methods train recommendation systems using supervised learning, leveraging large amounts of user behavioral history as training data [1,2,13]. ...
In large language models (LLM)-based recommendation systems (LLM-RSs), accurately predicting user preferences by leveraging the general knowledge of LLMs is possible without requiring extensive training data. By converting recommendation tasks into natural language inputs called prompts, LLM-RSs can efficiently solve issues that have been difficult to address due to data scarcity but are crucial in applications such as cold-start and cross-domain problems. However, when applying this in practice, selecting the prompt that matches tasks and data is essential. Although numerous prompts have been proposed in LLM-RSs and representing the target user in prompts significantly impacts recommendation accuracy, there are still no clear guidelines for selecting specific prompts. In this paper, we categorize and analyze prompts from previous research to establish practical prompt selection guidelines. Through 450 experiments with 90 prompts and five real-world datasets, we examined the relationship between prompts and dataset characteristics in recommendation accuracy. We found that no single prompt consistently outperforms others; thus, selecting prompts on the basis of dataset characteristics is crucial. Here, we propose a prompt selection method that achieves higher accuracy with minimal validation data. Because increasing the number of prompts to explore raises costs, we also introduce a cost-efficient strategy using high-performance and cost-efficient LLMs, significantly reducing exploration costs while maintaining high prediction accuracy. Our work offers valuable insights into the prompt selection, advancing accurate and efficient LLM-RSs.
... Graphs are ubiquitous in representing complex systems across diverse domains, from social networks 1 and biological systems 2 to transportation 3 and recommendation systems 4 . These structures, consisting of nodes and edges, capture relationships and interactions within the system. ...
Community structures are critical for understanding the mesoscopic organization of networks, bridging local and global patterns. While methods such as DeepWalk and node2vec capture local positional information through random walks, they fail to preserve community structures. Other approaches like modularized nonnegative matrix factorization and evolutionary algorithms address this gap but are computationally expensive and unsuitable for large-scale networks. To overcome these limitations, we propose Two Layer Walk (TLWalk), a novel graph embedding algorithm that incorporates hierarchical community structures. TLWalk balances intra- and inter-community relationships through a community-aware random walk mechanism without requiring additional parameters. Theoretical analysis demonstrates that TLWalk effectively mitigates locality bias. Experiments on benchmark datasets show that TLWalk outperforms state-of-the-art methods, achieving up to 3.2% accuracy gains for link prediction tasks. By encoding dense local and sparse global structures, TLWalk proves robust and scalable across diverse networks, offering an efficient solution for network analysis.
... • MF [48]: The traditional matrix factorization method. • BPR [84]: A classic recommendation model that optimizes matrix factorization using the Bayesian Personalized Ranking (BPR) objective function. ...
Recommender systems have become increasingly influential in shaping user behavior and decision-making, highlighting their growing impact in various domains. Meanwhile, the widespread adoption of machine learning models in recommender systems has raised significant concerns regarding user privacy and security. As compliance with privacy regulations becomes more critical, there is a pressing need to address the issue of recommendation unlearning, i.e., eliminating the memory of specific training data from the learned recommendation models. Despite its importance, traditional machine unlearning methods are ill-suited for recommendation unlearning due to the unique challenges posed by collaborative interactions and model parameters. This survey offers a comprehensive review of the latest advancements in recommendation unlearning, exploring the design principles, challenges, and methodologies associated with this emerging field. We provide a unified taxonomy that categorizes different recommendation unlearning approaches, followed by a summary of widely used benchmarks and metrics for evaluation. By reviewing the current state of research, this survey aims to guide the development of more efficient, scalable, and robust recommendation unlearning techniques. Furthermore, we identify open research questions in this field, which could pave the way for future innovations not only in recommendation unlearning but also in a broader range of unlearning tasks across different machine learning applications.
... Since users' activities on ecommerce websites are often infrequent, recommendation algorithms often face the sparsity problem, where transactions are too few as compared with users and items to build an effective model. From an algorithm perspective, dimension reduction methods, such as matrix factorization [37], have been incorporated to alleviate this problem. Another solution to this problem is to incorporate information to enrich user profiles. ...
Recommender systems are a critical component of e-commercewebsites. The rapid development of online social networking services provides an opportunity to explore social networks together with information used in traditional recommender systems, such as customer demographics, product characteristics, and transactions. It also provides more applications for recommender systems. To tackle this social network-based recommendation problem, previous studies generally built trust models in light of the social influence theory. This study inspects a spectrumof social network theories to systematicallymodel themultiple facets of a social network and infer user preferences. In order to effectively make use of these heterogonous theories, we take a kernel-based machine learning paradigm, design and select kernels describing individual similarities according to social network theories, and employ a non-linear multiple kernel learning algorithm to combine the kernels into a unified model. This design also enables us to consider multiple theories' interactions in assessing individual behaviors. We evaluate our proposed approach on a real-world movie review data set. The experiments show that our approach provides more accurate recommendations than trust-based methods and the collaborative filtering approach. Further analysis shows that kernels derived from contagion theory and homophily theory contribute a larger portion of the model.
... To account for these variations, researchers have incorporated user and item biases into matrix factorization models, enhancing performance [4]. Biased matrix factorization extends the basic model by adding two bias terms: one for the user (b i ) and one for the item (b j ). ...
In the field of Recommender Systems (RS), neural collaborative filtering represents a significant milestone by combining matrix factorization and deep neural networks to achieve promising results. Traditional methods like matrix factorization often rely on linear models, limiting their capability to capture complex interactions between users, items, and contexts. This limitation becomes particularly evident with high-dimensional datasets due to their inability to capture relationships among users, items, and contextual factors. Unsupervised learning and dimension reduction tasks utilize autoencoders, neural network-based models renowned for their capacity to encode and decode data. Autoencoders learn latent representations of inputs, reducing dataset size while capturing complex patterns and features. In this paper, we introduce a framework that combines neural contextual matrix factorization with autoencoders to predict user ratings for items. We provide a comprehensive overview of the framework's design and implementation. To evaluate its performance, we conduct experiments on various real-world datasets and compare the results against state-of-the-art approaches. We also extend the concept of conformal prediction to prediction rating and introduce a Conformal Prediction Rating (CPR). For RS, we define the nonconformity score, a key concept of conformal prediction, and demonstrate that it satisfies the exchangeability property.
... • Matrix Completion Problem (MCP) [14,24]: Consider the following MCP: ...
In this paper, we propose a Bregman proximal iteratively reweighted algorithm with extrapolation based on block coordinate update aimed at solving a class of optimization problems which is the sum of a smooth possibly nonconvex loss function and a general nonconvex regularizer with a separable structure. The proposed algorithm can be used to solve the regularization problem by employing an update strategy of the smoothing parameter in its smooth approximation model. When the extrapolation parameter satisfies certain conditions, the global convergence and local convergence rate are obtained by using the Kurdyka–Łojasiewicz (KL) property on the objective function. Numerical experiments are given to indicate the superiority of the proposed algorithm.
... This is a matrix factorization approach, of the kind commonly used in recommender systems to predict which items a user will like (Koren, Bell, and Volinsky 2009). In this case, exactly one factor is calculated for each note and each user, representing ideology or perspective along one axis. ...
Two substantial technological advances have reshaped the public square in recent decades: first with the advent of the internet and second with the recent introduction of large language models (LLMs). LLMs offer opportunities for a paradigm shift towards more decentralized, participatory online spaces that can be used to facilitate deliberative dialogues at scale, but also create risks of exacerbating societal schisms. Here, we explore four applications of LLMs to improve digital public squares: collective dialogue systems, bridging systems, community moderation, and proof-of-humanity systems. Building on the input from over 70 civil society experts and technologists, we argue that LLMs both afford promising opportunities to shift the paradigm for conversations at scale and pose distinct risks for digital public squares. We lay out an agenda for future research and investments in AI that will strengthen digital public squares and safeguard against potential misuses of AI.
... P ERSONALIZED recommender systems have been widely deployed on the Internet, aiming to provide essential web services to users and help them discover items of interest [1], [2]. Collaborative filtering (CF) [3] is widely utilized for generating recommendation lists using implicit user feedback. Lately, graph neural networks (GNNs) have been successful in capturing complex user-item interaction patterns through high-order representation learning [4], [5], [6]. ...
In the field of recommender systems, self-supervised learning has become an effective framework. In response to the noisy interaction behaviors in realworld scenarios, as well as the skewed distribution influenced by data sparsity and popularity bias, graph contrastive learning has been introduced as a powerful self-supervised method in collaborative filtering (CF) to learn enhanced user and item representations. Despite their success, neither heuristic manual enhancement methods nor the use of final node representations to construct contrastive pairs are sufficient to provide effective and rich self-supervised signals to regulate the training process. Therefore, the learned representations of users and items are either fragile or lack heuristic guidance. In light of this, we propose the Hierarchical multiview graph contrastive learning framework HMCF, which leverages the message passing mechanism at the layer level to introduce different granularity levels of view augmentation using supervised signals, thus better enhancing the CF paradigm. HMCF leverages rich, high-quality self-supervised signals from different granularity views for accurate contrastive optimization, helping to alleviate data sparsity and noise issues. It also explains the hierarchical topology and relative distances between nodes in the original graph. Comprehensive experiments on three public datasets shows that our model significantly outperforms the state-of-the-art baselines.
... where ݎ is the real rating, ܷ ் ܸ is the approximated rating, O and O are regularized parameters. PCA [22] analyzes high-dimensional data to simplify the useritem rating matrix, uncovering patterns and generating recommendations effectively with ܼ ൌ ܺ ڄ ܲ, where ܼ is the transformed data, X is the original data, and ܲ is the matrix of principal components (eigenvectors). SVD [23] decomposes a matrix into singular vectors and values, providing a rank-reduced approximation of the original matrix based on ܣ ൌ ܷܸܵ ் where ܷ is an ܯ ൈ ܰ matrix containing the left singular vectors, ܵ is an ܰ ൈ ܰ diagonal matrix with the singular values, and ܸ ் is an ܰ ൈ ܰ matrix with the right singular vectors. ...
People frequently struggle to make decisions when faced with a wider range of possibilities, especially when selecting a dining restaurant. To address this issue, a recommendation system can assist by analyzing user preferences and previous dining experiences to offer personalized suggestions. This research aims to develop a restaurant recommendation system for Malaysian customers using a machine-learning approach. The study focuses on Non-negative Matrix Factorization (NMF), Probability Matrix Factorization (PMF), Principal Component Analysis (PCA), and Singular Value Decomposition (SVD) approaches. Based on an analysis of 2,496 datasets gathered from the TripAdvisor platform, the findings revealed that the SVD method outperformed other approaches, achieving a Root Mean Square Error of 0.1166. This result positions SVD as the most suitable method for developing a restaurant recommendation system. The proposed system features a user-friendly interface built with Streamlit, allowing users to select their location and receive top restaurant suggestions. Additionally, users can view recommendations based on their past dining experiences. The system retrieves all reviews for the selected restaurants and converts them into a Term Frequency-Inverse Document Frequency (TF-IDF) matrix. Cosine similarity is then employed to measure the relevance of review content using the computed TF-IDF. Finally, the system also recommends similar restaurants based on the user’s chosen options, enhancing the overall dining experience.
... A practical example where this problem arises naturally is the Netflix problem: Users and movies are arranged in the rows and columns of a matrix M, respectively. Each entry M(i, j) represents the rating that user i gives to movie j [25]. In practice it is not possible to have all the ratings from all the users, and this is where matrix completion becomes important, assuming that the true matrix is approximately low-rank. ...
In this paper we generalize the 1-bit matrix completion problem to higher order tensors. We prove that when r=O(1) a bounded rank-r, order-d tensor T in can be estimated efficiently by only m=O(Nd) binary measurements by regularizing its max-qnorm and M-norm as surrogates for its rank. We prove that similar to the matrix case, i.e., when d=2, the sample complexity of recovering a low-rank tensor from 1-bit measurements of a subset of its entries is the same as recovering it from unquantized measurements. Moreover, we show the advantage of using 1-bit tensor completion over matricization both theoretically and numerically. Specifically, we show how the 1-bit measurement model can be used for context-aware recommender systems.
Motif 结构高阶相似度的文献推荐算法 * 陈 柳 郭宇红 (国际关系学院网络空间安全学院 北京 100091) 摘要: 【目的】 将协同过滤方法应用到文献推荐领域， 融入用户余弦相似度网络中 Motif 结构反映出的高阶相 似特征， 提高推荐的质量。 【方法】 通过用户收藏文献的行为信息和文献间的引用关系构建用户对文献的偏 好数据； 在基于用户-文献收藏行为信息的用户余弦相似度网络中， 利用网络中的子图-Motif 结构捕获高 阶相似度； 最后将用户余弦相似度和基于 Motif 结构的高阶相似度融入矩阵分解推荐算法中， 预测用户对文 献的偏好。 【结果】 相较于传统的矩阵分解推荐算法， 本文算法在 RMSE 和 MAE 指标上分别降低 0.0482 和 0.0379。 【局限】 未考虑文献的时间衰减性。 【结论】 本文算法降低了用户偏好预测误差， 提高了推荐质量。 关键词: 文献推荐 Motif 结构 用户高阶相似度 矩阵分解 分类号: TP391 G250
Motif 结构高阶相似度的文献推荐算法 * 陈 柳 郭宇红 (国际关系学院网络空间安全学院 北京 100091) 摘要: 【目的】 将协同过滤方法应用到文献推荐领域， 融入用户余弦相似度网络中 Motif 结构反映出的高阶相 似特征， 提高推荐的质量。 【方法】 通过用户收藏文献的行为信息和文献间的引用关系构建用户对文献的偏 好数据； 在基于用户-文献收藏行为信息的用户余弦相似度网络中， 利用网络中的子图-Motif 结构捕获高 阶相似度； 最后将用户余弦相似度和基于 Motif 结构的高阶相似度融入矩阵分解推荐算法中， 预测用户对文 献的偏好。 【结果】 相较于传统的矩阵分解推荐算法， 本文算法在 RMSE 和 MAE 指标上分别降低 0.0482 和 0.0379。 【局限】 未考虑文献的时间衰减性。 【结论】 本文算法降低了用户偏好预测误差， 提高了推荐质量。 关键词: 文献推荐 Motif 结构 用户高阶相似度 矩阵分解 分类号: TP391 G250
Accommodating the stakeholders’ preferences and mitigating their conflicts is a critical aspect of the Multi-Stakeholder Recommendation Systems (MSRS). Existing MSRS methods have addressed this by evaluating stakeholders’ utility gains. However, these methods fail to personalize the stakeholders’ goals, including providers’ aim to recommend their novel items and systems’ preference for fair recommendations, resulting in inequitable and unfair recommendations. Also, the optimization techniques of these methods neglect the stakeholders’ and item features while generating new recommendation solutions. Resolving these issues and optimizing the integration of conflicting multi-stakeholder utilities motivates this study to develop personalized objective functions focusing on relevance-visibility-based consumer preferences and novelty-fairness-based provider and system preferences. Next, we propose the Multi-Stakeholder Features-based Crossover (MSFCross) method to conserve the stakeholders’ preferences while generating new offspring by evaluating the cumulative probability of the gene and parent chromosome population. Finally, this study develops the MSFCross-based Preference Optimization (MSFCrossPO) framework to establish a balanced trade-off recommendation environment by optimizing the conflicting multi-stakeholder objectives. Comparative evaluation across five benchmark datasets validates the performance of MSFCrossPO employing MSFCross crossover scheme over baseline methods with the minimum improvement of 8.92 % in precision, 20.17 % in exposure, and 4.32 % and 10.94 % in consumer and provider satisfaction scores.
The growing interest in applications that generate playlists automatically per a user's mood highlights the need for new music technologies. This paper describes a web-based system that simultaneously uses machine learning techniques to recognize mood from text and facial features and generate an intelligent playlist. The application was developed on a React.js framework using Vite and Tailwind CSS and integrated Google Firebase in the back end of the application to make it a robust, reliable, and end-user-focused application. The paper also discusses the approach, design, system architecture, development, and evaluation of the system, which can change the way we perceive and use music streaming platforms. The evolution of music streaming platforms has transformed the way listeners interact with and consume music. With an overwhelming amount of music available at their fingertips, users often struggle to find tracks that resonate with their current mood or activity. Traditional playlist curation methods, either manual or algorithmic, frequently fail to capture the dynamic and multifaceted nature of human emotions. "MoodSync VibeBox: AI-Curated Playlists Tailored to Your Vibe" addresses this challenge by harnessing the power of artificial intelligence to revolutionize music personalization. By combining advanced mood detection algorithms, real-time data analysis, and adaptive playlist generation, this system offers an intuitive and engaging experience. MoodSync VibeBox transforms music listening from a passive to an emotionally resonant activity, bridging the gap between user emotions and personalized content delivery. This study aims to design and implement MoodSync VibeBox, an AI-powered system that seamlessly curates music playlists tailored to an individual's mood and activity. This platform aspires to, utilize cutting-edge AI algorithms to analyze emotional states in real-time. Deliver a seamless user experience that minimizes manual effort while maximizing engagement. Establish a benchmark in mood-based content delivery, setting the foundation for future advancements in music personalization technologies.
Multimodal recommendation systems can learn users' preferences from existing user-item interactions as well as the semantics of multimodal data associated with items. Many existing methods model this through a multimodal user-item graph, approaching multimodal recommendation as a graph learning task. Graph Neural Networks (GNNs) have shown promising performance in this domain. Prior research has capitalized on GNNs' capability to capture neighborhood information within certain receptive fields (typically denoted by the number of hops, K) to enrich user and item semantics. We observe that the optimal receptive fields for GNNs can vary across different modalities. In this paper, we propose GNNs with Modality-Independent Receptive Fields, which employ separate GNNs with independent receptive fields for different modalities to enhance performance. Our results indicate that the optimal K for certain modalities on specific datasets can be as low as 1 or 2, which may restrict the GNNs' capacity to capture global information. To address this, we introduce a Sampling-based Global Transformer, which utilizes uniform global sampling to effectively integrate global information for GNNs. We conduct comprehensive experiments that demonstrate the superiority of our approach over existing methods. Our code is publicly available at https://github.com/CrawlScript/MIG-GT.
Predicting thermodynamic properties of mixtures is a cornerstone of chemical engineering, yet conventional group-contribution (GC) methods like modified UNIFAC (Dortmund) remain limited by incomplete tables of pair-interaction parameters. To address this, we present modified UNIFAC 2.0, a hybrid model that integrates a matrix completion method from machine learning into the GC framework, allowing for the simultaneous training of all pair-interaction parameters, including the prediction of parameters that cannot be fitted due to missing data. Utilizing an extensive training set of more than 500,000 experimental data for activity coefficients and excess enthalpies from the Dortmund Data Bank, modified UNIFAC 2.0 achieves improved accuracy compared to the latest published version of modified UNIFAC (Dortmund) while significantly expanding the predictive scope. Its flexible design allows updates with new experimental data or customizations for specific applications. The new model can easily be implemented in established simulation software with complete parameter tables readily available.
Singular value decomposition (SVD), a data-driven fluid informatics method, was used to extract features of flow-induced single droplet breakup. The behavior of single droplet breakup induced by shockwave backflow was captured with high spatio-temporal resolution, and a large dataset was constructed by organizing the time and spatial luminance values according to the Weber number and its variations. SVD was performed on the dataset for space and variation, extracting spatial and variation features. Spatial modes indicating the upstream and downstream sides of the gas-liquid interface, as well as modes indicating planes perpendicular and parallel to the flow, were obtained. The sharp drop in the contribution rate of the second mode corresponds to the induction time of single droplet breakup. SVD was then applied to the dataset for space and time to extract spatial and temporal features. The first-order mode revealed features of the parent droplet and fragments with opposite signs, while higher-order modes captured detailed changes during the breakup process. The second mode, which shows the transition of the parent droplet, ligament, and fine fragment configurations, was emphasized. The moment when the time-mode feature becomes zero coincides with the induction time of droplet breakup. These results demonstrate that applying SVD to single droplet breakup provides a consistent and accurate determination of the breakup induction time.
This article describes a unique visual pedagogical tool that displays appropriate empirical antimicrobial treatment choices for anticipated pathogens. This new tool, a probability-based radial decision tree (RDT), transforms data from a traditional hospital antibiogram into a format that can be used as an epidemiologic tool, a guide to empiric antimicrobial therapy and a robust educational tool for display of therapy options for pan-susceptible and drug-resistant isolates. As well-described resistance mechanisms serve the basis for the antimicrobial treatment choices in the RDT, use of this tool provides a means for both displaying treatment choices and explaining the rationale for such options in a logical and systematic manner. Teaching fellow pharmacists and pharmacy students with this tool provides both verbal and visual cues that may allow more efficient conveying of key concepts important to understanding options for treatment of select pathogens.
Recommender systems provide users with personalized suggestions for products or services. These systems often rely on Collaborating Filtering (CF), where past transactions are analyzed in order to establish connections between users and products. The two more successful approaches to CF are latent factor models, which directly profile both users and products, and neighborhood models, which analyze similarities between products or users. In this work we introduce some innovations to both approaches. The factor and neighborhood models can now be smoothly merged, thereby building a more accurate combined model. Further accuracy improvements are achieved by extending the models to exploit both explicit and implicit feedback by the users. The methods are tested on the Netflix data. Results are better than those previously published on that dataset. In addition, we suggest a new evaluation metric, which highlights the differences among methods, based on their performance at a top-K recommendation task.
Customer preferences for products are drifting over time. Product perception and popularity are constantly changing as new selection emerges. Similarly, customer inclinations are evolving, leading them to ever redefine their taste. Thus, modeling temporal dynamics should be a key when designing recommender systems or general customer preference models. However, this raises unique challenges. Within the eco-system intersecting multiple products and customers, many different characteristics are shifting simultaneously, while many of them influence each other and often those shifts are delicate and associated with a few data instances. This distinguishes the problem from concept drift explorations, where mostly a single concept is tracked. Classical time-window or instance-decay approaches cannot work, as they lose too much signal when discarding data instances. A more sensitive approach is required, which can make better distinctions between transient effects and long term patterns. The paradigm we offer is creating a model tracking the time changing behavior throughout the life span of the data. This allows us to exploit the relevant components of all data instances, while discarding only what is modeled as being irrelevant. Accordingly, we revamp two leading collaborative filtering recommendation approaches. Evaluation is made on a large movie rating dataset by Netflix. Results are encouraging and better than those previously reported on this dataset.
predicated on the belief that information filtering can be more effective when humans are involved in the filtering process. Tapestry was designed to support both content-based filtering and collaborative filtering, which entails people collaborating to help each other perform filtering by recording their reactions to documents they read. The reactions are called annotations; they can be accessed by other people’s filters. Tapestry is intended to handle any incoming stream of electronic documents and serves both as a mail filter and repository; its components are the indexer, document store, annotation store, filterer, little box, remailer, appraiser and reader/browser. Tapestry’s client/server architecture, its various components, and the Tapestry query language are described.