No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Music Personalization at Spotify
Abstract
Spotify is the world's largest on-demand music streaming company, with over 75 million active listeners choosing what to listen to among tens of millions songs. Discovery and personalization is a key part of the experience and critical to the success of the creator and consumer ecosystem. In this talk, we'll discuss the state of our current discovery approaches, such as the Discover Weekly playlist that has already streamed billions of new discoveries and Fresh Finds, a scalable platform for brand new music that focuses suggestions on the long end of the popularity tail. We'll discuss the technologies at scale necessary to distill the information about music from our listeners and the world at large we collect outside of Spotify -- with the massive amounts of user-item activity data we collect every day to create highly personalized music experiences. Entire teams at Spotify focus on understanding both the creator and listener through collaborative filtering, machine learning, DSP and NLP approaches -- we crawl the web for artist information, scan each note in every one of our millions of songs for acoustic signals, and model users' taste through a cluster analysis and in a latent space based on their historical and real-time listening patterns. The data generated by these analyses have ensured our discovery products are precise and help our users enjoy music and media across our entire catalog. We'll dive deep into the workings of Discover Weekly, our marquee personalized playlist which updates weekly and reached 1 billion streams within the first 10 weeks from its release. The technology behind Discover Weekly is powered by a scalable factor analysis of Spotify's over two billion user-generated playlists matched to each user's current listening behavior. We'll discuss its innovative genesis and the challenges and opportunities the system faces a year after its launch. We'll also discuss Spotify's home page, seen by each of our users, currently undergoing vast efforts around personalization to ensure each listener gets a targeted list of playlists, shows and music to select throughout their day. We'll discuss the various similarity metrics, ranking approaches and user modeling we're working on to increase precision and optimize for our users' happiness.
... For each song we retrieve metadata -genres, popularity, artist etc. -and song features -tempo, key, loudness etc. -from the Spotify API. All features and metadata were created by Spotify [3]. ...
... Similarity can be represented in many layers of abstraction from raw audio signals to similarities between lyrics, to song metadata. The popular music streaming service Spotify uses a combination of these methods to personalize user recommendation [3]. ...
... With these identifiers, we used Spotify API to retrieve features and metadata about each song. Song features were created by Spotify through their proprietary music information retrieval models [3]. The features and metadata we used are detailed in Table 1. ...
There are many reasons people listen to music, and the type of music is largely determined by what the listener may be doing while they listen. For example, one may listen to one type of music while commuting, another while exercising, and yet another while relaxing. Without access to the physiological state of the user, current music recommendation methods rely on collaborative filtering - recommending music based on what other similar users listen to - and content based filtering - recommending songs based on their similarities to songs the user already prefers. With the rise in popularity of smart devices and activity trackers, physiological context can be a new channel to inform music recommendations. We propose deep learning solutions for context aware recommendation and playlist generation. Specifically, we use variational autoencoders (VAEs) to create a song embedding. We then explore multi-task multi-layer perceptrons (MLPs) and Gaussian mixture models to recommend songs based on context. We generate artificial user data to train and test our models in online learning and supervised learning settings.
... E-Entertainment (Music, Movies, Games, Dating Apps) Platforms like Netflix and Spotify personalize content recommendations using a mix of CF, CBF, and hybrid approaches, employing deep learning and ML to tailor suggestions based on user interactions and contextual factors [438,439]. Netflix utilizes deep learning and a blend of CF and CBF to analyze users' interactions and viewing habits [439,73], while Spotify leverages ML and NLP, introducing systems like GNN for audiobooks to address data sparsity and enhance content discovery [440,209,441]. The video game industry, exemplified by STEAM, uses advanced models to offer personalized game suggestions [442], addressing broader implications through multi-stakeholder recommendations [432]. ...
... Publications by Industry in Recommendation Systems Industry Publications E-commerce/E-Business [105, 435, 227, 303, 207, 211, 103, 181, 127, 43, 436, 437] E-Entertainment (Music, Movies)[438,439,73,440,209,441,442,432,20,41] ...
Recommender Systems (RS) play an integral role in enhancing user experiences by providing personalized item suggestions. This survey reviews the progress in RS inclusively from 2017 to 2024, effectively connecting theoretical advances with practical applications. We explore the development from traditional RS techniques like content-based and collaborative filtering to advanced methods involving deep learning, graph-based models, reinforcement learning, and large language models. We also discuss specialized systems such as context-aware, review-based, and fairness-aware RS. The primary goal of this survey is to bridge theory with practice. It addresses challenges across various sectors, including e-commerce, healthcare, and finance, emphasizing the need for scalable, real-time, and trustworthy solutions. Through this survey, we promote stronger partnerships between academic research and industry practices. The insights offered by this survey aim to guide industry professionals in optimizing RS deployment and to inspire future research directions, especially in addressing emerging technological and societal trends
... Recommender systems are becoming increasingly important for music streaming services such as Apple Music, Deezer, or Spotify [6,24,46]. While these services provide access to ever-growing musical catalogs, their recommender systems prevent information overload problems by identifying the most relevant content to showcase to each user [3,45]. ...
... While these services provide access to ever-growing musical catalogs, their recommender systems prevent information overload problems by identifying the most relevant content to showcase to each user [3,45]. Recommender systems also enable users to discover new songs, albums, or artists they may like [24,46]. Overall, they are widely regarded as effective tools to improve the user experience and engagement on these services [3,5,38,62]. ...
Music streaming services often aim to recommend songs for users to extend the playlists they have created on these services. However, extending playlists while preserving their musical characteristics and matching user preferences remains a challenging task, commonly referred to as Automatic Playlist Continuation (APC). Besides, while these services often need to select the best songs to recommend in real-time and among large catalogs with millions of candidates, recent research on APC mainly focused on models with few scalability guarantees and evaluated on relatively small datasets. In this paper, we introduce a general framework to build scalable yet effective APC models for large-scale applications. Based on a represent-then-aggregate strategy, it ensures scalability by design while remaining flexible enough to incorporate a wide range of representation learning and sequence modeling techniques, e.g., based on Transformers. We demonstrate the relevance of this framework through in-depth experimental validation on Spotify's Million Playlist Dataset (MPD), the largest public dataset for APC. We also describe how, in 2022, we successfully leveraged this framework to improve APC in production on Deezer. We report results from a large-scale online A/B test on this service, emphasizing the practical impact of our approach in such a real-world application.
... In particular, several recent works emphasized the empirical effectiveness of latent models for collaborative filtering at addressing industrial-level challenges [63,114,164,339]. Analogously to the node embedding methods developed throughout this thesis, these models aim to directly learn vector space representations, i.e., embeddings of users and items where proximity should reflect user preferences, typically via the factorization of a user-item interaction matrix [139,195,196] or with neural networks architectures processing usage data [63,260,372]. ...
... Besides, one can enrich such systems by explicitly asking new users to rate items from the catalog through interview processes, leading to hybrid models based on preferences and side information [118,263,331]. On the industry side, Netflix [114] and Spotify [164] are famous examples of services implementing such onboarding session for new users. As explained in Section 12.3, 212 ...
Les autoencodeurs de graphes (GAE) et les autoencodeurs variationnels de graphes (VGAE) se sont imposés comme deux puissants groupes de méthodes permettant de construire des représentations vectorielles des nœuds d'un graphe de manière non supervisée, avec des applications à divers problèmes d'apprentissage tels que la prédiction de liens manquants et la détection de communautés de nœuds. Néanmoins, au début de ce projet de thèse, les GAE et VGAE souffraient de limitations majeures. Ces dernières entravaient l'utilisation de ces modèles dans le cadre d'applications industrielles. Dans cette thèse, nous présentons plusieurs contributions permettant d'améliorer les GAE et VGAE afin de faciliter de telles utilisations.
Tout d'abord, nous proposons deux stratégies permettant de surmonter les problèmes de passage à l'échelle des GAE et VGAE, et d'entraîner ces modèles sur des graphes ayant des millions de nœuds et d'arêtes. Ces stratégies exploitent respectivement des techniques de dégénérescence de graphes et de décodage stochastique de sous-graphes. Par ailleurs, nous présentons nos GAE et VGAE inspirés de la gravité (de l'anglais Gravity-Inspired GAE and VGAE), qui constituent les premières extensions de ces modèles destinées aux graphes dirigés, qui sont omniprésents dans les applications industrielles. Nous étudions également des extensions destinées aux graphes dynamiques. En outre, nous démontrons que les GAE et VGAE existants sont souvent inutilement complexes, et nous proposons donc de les simplifier en ayant recours à des encodeurs linéaires. Enfin, nous présentons nos GAE et VGAE informés par la modularité (de l'anglais Modularity-Aware GAE and VGAE), qui permettent d'améliorer la détection de communautés de nœuds, tout en préservant de bonnes performances pour la prédiction de liens manquants.
Dans la dernière partie de cette thèse, nous évaluons nos méthodes sur plusieurs graphes extraits du service de streaming musical Deezer. Nous nous concentrons sur des problèmes de recommandation musicale à partir de graphes. En particulier, nous montrons que nos méthodes permettent d'améliorer la détection de communautés d'entités musicales à recommander aux mêmes utilisateurs, mais aussi de mieux classer des artistes similaires dans un contexte de démarrage à froid (de l'anglais cold start), et enfin de mieux modéliser la perception des genres musicaux à travers différentes cultures. Pour terminer, nous présentons également deux autres modèles, récemment déployés en production chez Deezer afin de recommander de la musique à des millions d'utilisateurs. Bien qu'étant moins directement liés aux GAE et VGAE, ils fournissent un point de vue complémentaire sur des sujets de recommandation musicale connexes à ceux étudiés précédemment.
... One-time search engines treat each query independently. Life-time search engines consider data about the user from the user's first search up to the present, to improve their results [27,35,22,12,19,7]. ...
... One approach is to improve the search results using additional data about the user. For example, Hersh et al. [16] utilize implicit user data to improve results, while [19] uses meta-data from other sources of information about the user. Other search engines take advantage of both explicit and implicit data (hybrid) from the user [12,39]. ...
Finding relevant research literature in online databases is a familiar challenge to all researchers. General search approaches trying to tackle this challenge fall into two groups: one-time search and life-time search. We observe that both approaches ignore unique attributes of the research domain and are affected by concept drift. We posit that in searching for research papers, a combination of a life-time search engine with an explicitly-provided context (project) provides a solution to the concept drift problem. We developed and deployed a project-based meta-search engine for research papers called Rivendell. Using Rivendell, we conducted experiments with 199 subjects, comparing project-based search performance to one-time and life-time search engines, revealing an improvement of up to 12.8 percent in project-based search compared to life-time search.
... In particular, several recent works emphasized the empirical effectiveness of latent models for collaborative filtering at addressing industrial-level challenges [63,114,164,339]. Analogously to the node embedding methods developed throughout this thesis, these models aim to directly learn vector space representations, i.e., embeddings of users and items where proximity should reflect user preferences, typically via the factorization of a user-item interaction matrix [139,195,196] or with neural networks architectures processing usage data [63,260,372]. ...
... Besides, one can enrich such systems by explicitly asking new users to rate items from the catalog through interview processes, leading to hybrid models based on preferences and side information [118,263,331]. On the industry side, Netflix [114] and Spotify [164] are famous examples of services implementing such onboarding session for new users. As explained in Section 12.3, 212 ...
Graph autoencoders (GAE) and variational graph autoencoders (VGAE) emerged as two powerful groups of unsupervised node embedding methods, with various applications to graph-based machine learning problems such as link prediction and community detection. Nonetheless, at the beginning of this Ph.D. project, GAE and VGAE models were also suffering from key limitations, preventing them from being adopted in the industry. In this thesis, we present several contributions to improve these models, with the general aim of facilitating their use to address industrial-level problems involving graph representations. Firstly, we propose two strategies to overcome the scalability issues of previous GAE and VGAE models, permitting to effectively train these models on large graphs with millions of nodes and edges. These strategies leverage graph degeneracy and stochastic subgraph decoding techniques, respectively. Besides, we introduce Gravity-Inspired GAE and VGAE, providing the first extensions of these models for directed graphs, that are ubiquitous in industrial applications. We also consider extensions of GAE and VGAE models for dynamic graphs. Furthermore, we argue that GAE and VGAE models are often unnecessarily complex, and we propose to simplify them by leveraging linear encoders. Lastly, we introduce Modularity-Aware GAE and VGAE to improve community detection on graphs, while jointly preserving good performances on link prediction. In the last part of this thesis, we evaluate our methods on several graphs extracted from the music streaming service Deezer. We put the emphasis on graph-based music recommendation problems. In particular, we show that our methods can improve the detection of communities of similar musical items to recommend to users, that they can effectively rank similar artists in a cold start setting, and that they permit modeling the music genre perception across cultures.
... In particular, several recent works emphasized the empirical effectiveness of latent models for collaborative filtering at addressing industrial-level challenges [63,114,164,339]. Analogously to the node embedding methods developed throughout this thesis, these models aim to directly learn vector space representations, i.e., embeddings of users and items where proximity should reflect user preferences, typically via the factorization of a user-item interaction matrix [139,195,196] or with neural networks architectures processing usage data [63,260,372]. ...
... Besides, one can enrich such systems by explicitly asking new users to rate items from the catalog through interview processes, leading to hybrid models based on preferences and side information [118,263,331]. On the industry side, Netflix [114] and Spotify [164] are famous examples of services implementing such onboarding session for new users. As explained in Section 12.3, 212 ...
Graph autoencoders (GAE) and variational graph autoencoders (VGAE) emerged as two powerful groups of unsupervised node embedding methods, with various applications to graph-based machine learning problems such as link prediction and community detection. Nonetheless, at the beginning of this Ph.D. project, GAE and VGAE models were also suffering from key limitations, preventing them from being adopted in the industry. In this thesis, we present several contributions to improve these models, with the general aim of facilitating their use to address industrial-level problems involving graph representations.Firstly, we propose two strategies to overcome the scalability issues of previous GAE and VGAE models, permitting to effectively train these models on large graphs with millions of nodes and edges. These strategies leverage graph degeneracy and stochastic subgraph decoding techniques, respectively. Besides, we introduce Gravity-Inspired GAE and VGAE, providing the first extensions of these models for directed graphs, that are ubiquitous in industrial applications. We also consider extensions of GAE and VGAE models for dynamic graphs. Furthermore, we argue that GAE and VGAE models are often unnecessarily complex, and we propose to simplify them by leveraging linear encoders. Lastly, we introduce Modularity-Aware GAE and VGAE to improve community detection on graphs, while jointly preserving good performances on link prediction.In the last part of this thesis, we evaluate our methods on several graphsextracted from the music streaming service Deezer. We put the emphasis on graph-based music recommendation problems. In particular, we show that our methods can improve the detection of communities of similar musical items to recommend to users, that they can effectively rank similar artists in a cold start setting, and that they permit modeling the music genre perception across cultures. At the end of this thesis, we present two additional models, recently deployed in production on the Deezer service to recommend music to millions of users. While being less directly linked to GAE and VGAE models, they provide a complementary perspective on music recommendation topics related to the ones we previously studied.
... To mitigate these constraints, researchers employed hybrid recommendation systems that integrate collaborative filtering with content-based methods. Although these hybrid systems represented progress, they remained inadequate in encapsulating the intricacies of human preferences, particularly as music consumption has evolved to be increasingly multi-modal, encompassing audio characteristics, metadata, and user interactions [6][7][8]. As a result, deep learning has emerged as a viable method to address these difficulties. ...
This paper provides a thorough examination of the utilisation of deep learning in music recommendation systems, which have transformed consumer discovery and engagement with music on streaming platforms. Scalability challenges and the cold-start problem are among the constraints that conventional recommendation methods, such as content-based filtering and collaborative filtering, encounter, which hinder their ability to deliver personalised recommendations that are effective. The processing of multi-modal and sequential data is significantly improved by deep learning methodologies, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Autoencoders. This results in more precise and contextually pertinent music recommendations. Moreover, hybrid models that amalgamate deep learning with conventional techniques augment recommendation precision by synthesising user interaction data with acoustic characteristics. This paper examines essential performance metrics employed in the assessment of music recommendation systems, including precision, recall, F1-score, and Mean Reciprocal Rank (MRR). It also tackles issues including computational complexity, bias, and ethical dilemmas pertaining to data privacy.
... By analyzing watching trends, Netflix may generate personalized content recommendations, facilitating viewers' discovery of media that captivates their interest. Spotify [4] employs analogous algorithms to suggest music tracks and playlists that correspond with the user's preferences in genres, artists, and songs. There are several ways to build recommender systems. ...
Recommender systems are essential for providing personalized content across various platforms. However, traditional systems often struggle with limited information, known as the cold start problem, and with accurately interpreting a user's comprehensive preferences, referred to as context. The proposed study, CD-LLMCARS (Cross-Domain fine-tuned Large Language Model for Context-Aware Recommender Systems), presents a novel approach to addressing these issues. CD-LLMCARS leverages the substantial capabilities of the Large Language Model (LLM) Llama 2. Fine-tuning Llama 2 with information from multiple domains can enhance the generation of contextually relevant recommendations that align with a user's preferences in areas such as movies, music, books, and CDs. Techniques such as Low-Rank Adaptation (LoRA) and Half Precision Training (FP16) are both effective and resource-efficient, allowing CD-LLMCARS to perform optimally in cold start scenarios. Extensive testing of CD-LLMCARS indicates outstanding accuracy, particularly in challenging scenarios characterized by limited user history data relevant to the cold start problem. CD-LLMCARS offers precise and pertinent recommendations to users, effectively mitigating the limitations of traditional recommender systems.
... Request permissions from permissions@acm.org. RecSys '24, October [14][15][16][17][18]2024, Bari, Italy academic [1,28,61,64] and industrial researchers [12,27,37], who have studied various methods associated with personalization. ...
Bayesian Personalized Ranking (BPR), a collaborative filtering approach based on matrix factorization, frequently serves as a benchmark for recommender systems research. However, numerous studies often overlook the nuances of BPR implementation, claiming that it performs worse than newly proposed methods across various tasks. In this paper, we thoroughly examine the features of the BPR model, indicating their impact on its performance, and investigate open-source BPR implementations. Our analysis reveals inconsistencies between these implementations and the original BPR paper, leading to a significant decrease in performance of up to 50% for specific implementations. Furthermore, through extensive experiments on real-world datasets under modern evaluation settings, we demonstrate that with proper tuning of its hyperparameters, the BPR model can achieve performance levels close to state-of-the-art methods on the top-n recommendation tasks and even outperform them on specific datasets. Specifically, on the Million Song Dataset, the BPR model with hyperparameters tuning statistically significantly outperforms Mult-VAE by 10% in NDCG@100 with binary relevance function.
... Recommender systems are essential to online platforms providing access to large catalogs, such as music streaming services [26,59]. They mitigate information overload by identifying the most relevant content to showcase to users, e.g., personalized song selections for a music streaming service [47]. ...
Music streaming services often leverage sequential recommender systems to predict the best music to showcase to users based on past sequences of listening sessions. Nonetheless, most sequential recommendation methods ignore or insufficiently account for repetitive behaviors. This is a crucial limitation for music recommendation, as repeatedly listening to the same song over time is a common phenomenon that can even change the way users perceive this song. In this paper, we introduce PISA (Psychology-Informed Session embedding using ACT-R), a session-level sequential recommender system that overcomes this limitation. PISA employs a Transformer architecture learning embedding representations of listening sessions and users using attention mechanisms inspired by Anderson's ACT-R (Adaptive Control of Thought-Rational), a cognitive architecture modeling human information access and memory dynamics. This approach enables us to capture dynamic and repetitive patterns from user behaviors, allowing us to effectively predict the songs they will listen to in subsequent sessions, whether they are repeated or new ones. We demonstrate the empirical relevance of PISA using both publicly available listening data from Last.fm and proprietary data from Deezer, a global music streaming service, confirming the critical importance of repetition modeling for sequential listening session recommendation. Along with this paper, we publicly release our proprietary dataset to foster future research in this field, as well as the source code of PISA to facilitate its future use.
... Recommender systems are essential for music streaming services like Apple Music, Deezer, and Spotify [9,26,38,39]. They help mitigate information overload problems by showcasing the most relevant content for each user, within large catalogs of millions of songs, albums, and artists [7,19,23,32,39]. ...
This paper examines the influence of recommender systems on local music representation, discussing prior findings from an empirical study on the LFM-2b public dataset. This prior study argued that different recommender systems exhibit algorithmic biases shifting music consumption either towards or against local content. However, LFM-2b users do not reflect the diverse audience of music streaming services. To assess the robustness of this study's conclusions, we conduct a comparative analysis using proprietary listening data from a global music streaming service, which we publicly release alongside this paper. We observe significant differences in local music consumption patterns between our dataset and LFM-2b, suggesting that caution should be exercised when drawing conclusions on local music based solely on LFM-2b. Moreover, we show that the algorithmic biases exhibited in the original work vary in our dataset, and that several unexplored model parameters can significantly influence these biases and affect the study's conclusion on both datasets. Finally, we discuss the complexity of accurately labeling local music, emphasizing the risk of misleading conclusions due to unreliable, biased, or incomplete labels. To encourage further research and ensure reproducibility, we have publicly shared our dataset and code.
... Machine learning is a study attempting to extract knowledge from available data using computer programs that can learn and get smarter automatically based on experience [36] [37]. Currently, the application of machine learning can be found in various activities in everyday life, such as recommendations for goods in Amazon e-commerce services [38], recommendations on the music streaming platform Spotify [39], and recommendations in education assessment [40][41] [42]In bioinformatics, machine learning has been widely used to solve problems in various areas, including genomics, proteomics, systems biology, evolution, microarrays, and text mining [43][44] [45]. The application of machine learning in each case handles the different characteristics of the input data. ...
Classifying plant species within the Liliaceae and Amaryllidaceae families presents inherent challenges due to the complex genetic diversity and overlapping morphological traits among species. This study explores the difficulties in accurate classification by comparing 11 supervised learning algorithms applied to DNA barcode data, aiming to enhance the precision of species family classification in these taxonomically intricate plant families. The ribulose-1,5-bisphosphate carboxylase-oxygenase large sub-unit (rbcL) gene, selected as a DNA barcode locus for plants, is used to represent species within the Amaryllidaceae and Liliaceae families. The experimental results demonstrate that nearly all tested models achieve accurate species classification into the appropriate families, with an accuracy rate exceeding 97%, except for the Naïve Bayes model. Regarding computational time, the Random Forest model requires significantly more time for training than other models. Regarding memory usage, the Least Squares Support Vector Machine with a polynomial kernel, and Regularized Logistic Regression consume more memory than other models. These machine learning models exhibit strong concordance with NCBI's classifications when predicting families using the test dataset, effectively categorizing species into the Amaryllidaceae and Liliaceae families.
... These approaches have successfully altered the simple popularity-based item list traditionally used by many web applications like Amazon, Spotify, and Netflix. Also, such recommendation strategies focus only on accuracy in the design and evaluation of recommendation systems [26,31,49]. Earlier approaches suffer from drawbacks like redundancy where recommending items based on the user's past interaction history, limiting their interaction domains, i.e., a reduction of diversity in the model. ...
Many approaches based on Graph Neural Networks (GNNs) have been proposed to identify relationships between users and items while modelling user preferences with significant improvements in recommendation quality. Besides accuracy, diversity in recommendation is often a desirable property for a better user experience in a real-world application. Recently many recommendation techniques based on heterogeneous information networks have been drawing attention to improvement in diversity. However, most such algorithms use re-ranking approaches or diversity regularization (ensemble learning) in a heterogeneous graph network. These approaches often compromise with accuracy to include diversity in the recommendation. The author proposed a novel technique involving both diversity and accuracy at the same time for recommendation generation. Our approach uses implicit user information to generate a low-dimensional embedding representation for each node. The model also includes derived user features for diversity to train the model for diversified recommendation generation. The proposed model iteratively finds infrequently recommended yet relevant items, adds them to the users’ final recommendation lists, and balances the accuracy diversity tradeoff. Extensive experiments on real-world datasets demonstrate the effectiveness of our proposed model, Diverse Heterogeneous Node Embedding Model for Recommendation (Div-HetNEMRec), for diverse recommendations with substantially better coverage and reasonably good improvement in accuracy over the state-of-the-art techniques.
... Since its introduction for the 1 st time in the 1950s, artificial intelligence (AI) has captured the imagination of one and all. [1] With continuous technological developments and AI becoming a part of our everyday lives through various applications, [2,3] the fascination with AI has only increased further. AI has become a part of the healthcare sector, in the form of advanced machinery [4,5] and various software. ...
Artificial intelligence (AI) is taking its grasp over health-care system and ophthalmology as one of the most dynamic streams is largely influenced by AI. AI over the past few decades has made a huge impact in the bailiwick of oculoplasty. AI-based imaging softwares have made easier the diagnosis and management of several orbital and eyelid pathologies by its accuracy and reproducibility. AI also has made possible real-time tracking of deep orbital structures through navigation-guided technologies which have made orbital surgeries safer and easier. This article is a meta-analysis of several articles which have discussed applications and impact of AI-based software in diagnosis and management planning of periorbital and eyelid pathologies and also articles on navigation-guided orbital surgeries.
... Here we propose a new recommendation scenario, called taste cluster recommendation, which means users are recommended with a bundle of similar items, and these items can be described with few tags (or features). We find this task is ubiquitous in real-world applications, like playlist recommendation in Spotify [14]. Current solutions to tackle this problem typically rely on manual selection by editors to guarantee the selected items are similar in tags (or other explicit features). ...
Collaborative Filtering (CF) is a widely used and effective technique for recommender systems. In recent decades, there have been significant advancements in latent embedding-based CF methods for improved accuracy, such as matrix factorization, neural collaborative filtering, and LightGCN. However, the explainability of these models has not been fully explored. Adding explainability to recommendation models can not only increase trust in the decisionmaking process, but also have multiple benefits such as providing persuasive explanations for item recommendations, creating explicit profiles for users and items, and assisting item producers in design improvements. In this paper, we propose a neat and effective Explainable Collaborative Filtering (ECF) model that leverages interpretable cluster learning to achieve the two most demanding objectives: (1) Precise - the model should not compromise accuracy in the pursuit of explainability; and (2) Self-explainable - the model's explanations should truly reflect its decision-making process, not generated from post-hoc methods. The core of ECF is mining taste clusters from user-item interactions and item profiles.We map each user and item to a sparse set of taste clusters, and taste clusters are distinguished by a few representative tags. The user-item preference, users/items' cluster affiliations, and the generation of taste clusters are jointly optimized in an end-to-end manner. Additionally, we introduce a forest mechanism to ensure the model's accuracy, explainability, and diversity. To comprehensively evaluate the explainability quality of taste clusters, we design several quantitative metrics, including in-cluster item coverage, tag utilization, silhouette, and informativeness. Our model's effectiveness is demonstrated through extensive experiments on three real-world datasets.
... Other field settings might be able to capture such dynamics for more endings as well. The popular music streaming service Spotify offers its users custom-curated playlists, including "Release Radar" and "Discover Weekly" playlists filled with novel songs that users are likely to enjoy, as well as "On Repeat" and "Repeat Rewind" playlists filled with familiar songs based on users' own past song choices (Jacobson et al., 2016). Such offerings parallel our own Experiment 7 in ways that may be fruitfully tapped by future field research. ...
... Spotify is the world's largest on-demand music streaming company, with over 75 million active listeners choosing what to listen to among tens of millions of songs. Discovery and personalization are key parts of the experience and critical to the success of the creator and consumer ecosystem [5]. Spotify clears users' playlists every Monday and suggests songs the company thinks they'll like. ...
With the change in people's music consumption patterns, music streaming media enterprises have attracted more and more attention. Spotify is one of the largest music streaming media enterprises at present, and its rapid development has been attracting attention. As the market becomes increasingly competitive, Spotify's single business model is increasingly limited. Through the combination of qualitative and quantitative research, this paper concludes that the main problems of Spotify's current business model are as follows: firstly, it cannot produce content, and acquiring music playing rights only by buying the copyright of record companies not only consumes a lot of money but also has uncertainties. Second, only relying on a single advertising and subscription fees revenue model has been unable to meet the needs of enterprise management costs. Third, as the competition among music-streaming services intensifies, Spotify must be able to provide users with music that is more satisfying to their tastes. Therefore, to solve these problems in the future, Spotify can focus on creating its own record company, upgrading its algorithm, and developing its own hardware products to enrich its business model.
... Fourth and final: Context-aware computing. Context-aware recommender system (CARS) predicted 15 years ago as a part of recommender system based on the context of the action (time, location) [Adomavicius and Tuzhilin, 2011;Panniello and al., 2014] are implemented by the major companies (such as Netflix [Gomez-Uribe and Hunt, 2015], Amazon [Smith and Linden, 2017] and Spotify [Jacobson and al., 2016]) to solve the problem of overwhelming users by the huge amount of available information. The information to predict customers' needs (contextual information) could be acquired by explicit (when the user enters contextual information directly), implicit (by observing the user's behavior) and Inferred (using statistical or data mining methods) acquisition [del Carmen Rodríguez-Hernández and Ilarri, 2021]. ...
Marketing personalization is an essential tool for professionals to improve efficiency of marketing campaigns in digital channels, especially creating individual offers for customers. Artificial intelligence seems to be the magic wand of marketing personalization by providing unlimited opportunity to collect segments data and find the most appropriate content for different types of customers. The current paper is a systematic literature review (2017-2022) about Machine Learning and marketing communication between customer and firm in the real estate industry. Synthesis presents the state of the art as well as future resaerch orientations for academics.
... In industries such as retail, e-commerce, media apps, or even healthcare, recommendation system models (see, e.g., (Ricci, Rokach, and Shapira 2015;Burke, Felfernig, and Göker 2011)) are critical to customer retention. Corporations like Netflix, Spotify, Amazon, etc., use sophisticated collaborative filtering and content-based recommendation systems for video, song, and/or product recommendations (Gomez-Uribe and Hunt 2016; Amatriain and Basilico 2015;Jacobson et al. 2016;Smith and Linden 2017;Wang et al. 2014). For a recent overview of recommender systems in the healthcare domain see, e.g., (Tran et al. 2021) and the references therein. ...
This effort is focused on examining the behavior of reinforcement learning systems in personalization environments and detailing the differences in policy entropy associated with the type of learning algorithm utilized. We demonstrate that Policy Optimization agents often possess low-entropy policies during training, which in practice results in agents prioritizing certain actions and avoiding others. Conversely, we also show that Q-Learning agents are far less susceptible to such behavior and generally maintain high-entropy policies throughout training, which is often preferable in real-world applications. We provide a wide range of numerical experiments as well as theoretical justification to show that these differences in entropy are due to the type of learning being employed.
... With greater availability of music from more access-based consumption [2,27], music recommendations-one form of music sharing-continues to be important. 1 Given the sheer amount of music that users have access to, streaming platforms have focused their efforts on algorithmic personalized recommendations [35]. Yet research shows that not all recommendations are equal-music shared by people are oftentimes perceived more positively compared to those from systems [37,48]. ...
Music sharing is a common social activity that people have long engaged in, from gifting mixtapes to sharing music links. Our practices around sharing music have shifted markedly with the advent of streaming music platforms and social media, and it has remained an important part of our social fabric. Yet there is a dearth of research on how people share music today, and our understanding of attitudes and practices of sharing music across cultures is even more lacking. To understand how people across cultures engage in music sharing, we have conducted interviews with 32 participants from two cultures: South Korea and United States. Through qualitative analysis, we found largely three reasons why people share music, types of music shared, strategy factors considered when sharing music, outcomes achieved, and challenges people experience when sharing music. We present a framework of music sharing that visualizes these components of the music sharing process. From these results, we identify similarities and differences that emerged. We derive design implications for music sharing platforms including providing varied avenues for feedback on shared music, motivating users to share more, and helping users to better manage shared music.
... Kurt et al. conducted a personalized music study by the Spotify application in the paper. They have recommended an appeal to user pleasure through real-time resting molds [14]. In [15], a Spotify application uses digital ads in which the content for the users is investigated. ...
This research aims to analyze the effect of feature selection on the accuracy of music popularity classification using machine learning algorithms. The data of Spotify, the most used music listening platform today, was used in the research. In the feature selection stage, features with low correlation were removed from the dataset using the filter feature selection method. Machine learning algorithms using all features produced 95.15% accuracy, while machine learning algorithms using features selected by feature selection produced 95.14% accuracy. The features selected by feature selection were sufficient for classification of popularity in established algorithms. In addition, this dataset contains fewer features, so the computation time is shorter. The reason why Big O time complexity is lower than models constructed without feature selection is that the number of features, which is the most important parameter in time complexity, is low. The statistical analysis was performed on the pre-processed data and meaningful information was produced from the data using machine learning algorithms.
... Other field settings might be able to capture such dynamics for more endings as well. The popular music streaming service Spotify offers its users custom-curated playlists, including "Release Radar" and "Discover Weekly" playlists filled with novel songs that users are likely to enjoy, as well as "On Repeat" and "Repeat Rewind" playlists filled with familiar songs based on users' own past song choices (Jacobson et al., 2016). Such offerings parallel our own Experiment 7 in ways that may be fruitfully tapped by future field research. ...
People fill their free time by choosing between hedonic activities that are new and exciting (e.g., exploring a buzzed-about restaurant) versus old and familiar (e.g., revisiting the same old spot). The dominant psychological assumption is that people will prefer novelty, holding constant factors like cost, availability, and convenience between acquiring such options ("variety is the spice of life"). Eight preregistered experiments (total N = 5,889) reveal that people's attraction to novelty depends, at least in part, on their temporal context-namely, on perceived endings. As participants faced a shrinking window of opportunity to enjoy a general category of experience (even merely temporarily; e.g., eating one's last dessert before starting a diet), their hedonic preferences shifted away from new and exciting options and toward old favorites. This relative shift emerged across many domains (e.g., food, travel, music), situations (e.g., impending New Year's resolutions, COVID-19 shutdowns), and consequential behaviors (e.g., choices with financial stakes). Using both moderation and mediation approaches, we found that perceived endings increase the preference for familiarity because they increase people's desire to ensure a personally meaningful experience on which to end, and returning to old favorites is typically more meaningful than exploring novelty. Endings increased participants' preference for familiarity even when it meant sacrificing other desirable attributes (e.g., exciting stimulation). Together, these findings advance and bridge research on hedonic preferences, time and timing, and the motivational effects of change. Variety may be the "spice of life," but familiarity may be the spice of life's endings. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
... This paper relates to several strands of literature. Personalized recommendation systems have been studied intensively in entertainment (Davidson et al., 2010;Gomez-Uribe and Hunt, 2015;Jacobson et al., 2016;Holtz et al., 2020) and in retail shopping (Linden et al., 2003;Sharma et al., 2015;Smith and Linden, 2017;Greenstein-Messica and Rokach, 2018;Ursu, 2018). For example, in the entertainment context and using a similar approach to our paper, (Holtz et al., 2020) show that personalized recommendations increase consumption of podcasts on Spotify. ...
We study the impact of personalized content recommendations on the usage of an educational app for children. In a randomized controlled trial, we show that the introduction of personalized recommendations increases the consumption of content in the personalized section of the app by approximately 60% and that the overall app usage increases by 14%, compared to the baseline system of stories selected by content editors for all students. The magnitude of individual gains from personalized content increases with the amount of data available about a student and with preferences for niche content: heavy users with long histories of content interactions who prefer niche content benefit more than infrequent, newer users who like popular content. To facilitate the diffusion of personalized recommendation systems, we provide a framework for using offline data to develop such a system.
... Some transfer learning-based methods [2,25,35] alleviate the cold-start problem by transferring well-learned representations of overlapped objects from the source domain to the target domain. The active learning scheme [20,43,64] explicitly encourages new users to rate items from the catalog through various interview processes with the extra costs or budgets. Additionally, GNN-based models [17,53,54,58] have been developed for recommendation, which utilizes user-item bipartite graph to capture high-order collaborative signal. ...
The cold-start problem is a long-standing challenge in recommender systems due to the lack of user-item interactions, which significantly hurts the recommendation effect over new users and items. Recently, meta-learning based methods attempt to learn globally shared prior knowledge across all users, which can be rapidly adapted to new users and items with very few interactions. Though with significant performance improvement, the globally shared parameter may lead to local optimum. Besides, they are oblivious to the inherent information and feature interactions existing in the new users and items, which are critical in cold-start scenarios. In this paper, we propose a Task aligned Meta-learning based Augmented Graph (TMAG) to address cold-start recommendation. Specifically, a fine-grained task aligned constructor is proposed to cluster similar users and divide tasks for meta-learning, enabling consistent optimization direction. Besides, an augmented graph neural network with two graph enhanced approaches is designed to alleviate data sparsity and capture the high-order user-item interactions. We validate our approach on three real-world datasets in various cold-start scenarios, showing the superiority of TMAG over state-of-the-art methods for cold-start recommendation.
... Recommender systems are an essential part of music streaming services [1,13,20,21]. They allow users to discover new songs or artists they may like within large music catalogs, and they are known to improve the overall user experience on these services [5,22]. ...
The music streaming service Deezer extensively relies on its Flow algorithm, which generates personalized radio-style playlists of songs, to help users discover musical content. Nonetheless, despite promising results over the past years, Flow used to ignore the moods of users when providing recommendations. In this paper, we present Flow Moods, an improved version of Flow that addresses this limitation. Flow Moods leverages collaborative filtering, audio content analysis, and mood annotations from professional music curators to generate personalized mood-specific playlists at scale. We detail the motivations, the development, and the deployment of this system on Deezer. Since its release in 2021, Flow Moods has been recommending music by moods to millions of users every day.
... In other words, the current naïve approach to online moderation neglects individual differences and the advantages of personalization. Indeed, personalization has already proved valuable in several online domains, such as advertising, music and video streaming services, and the improvement of health-related behavior via apps [12,26]. By taking inspiration from medicine -a field with which online moderation shares many commonalities-we observe that current generic interventions are deployed as a sort of universal cure to treat the ailments of all online users, instead of following the virtuous examples of personalized medicine where patients affected by a disease receive personalized treatment based on their condition, individual characteristics, environment, and behavior. ...
Current online moderation follows a one-size-fits-all approach, where each intervention is applied in the same way to all users. This naive approach is challenged by established socio-behavioral theories and by recent empirical results that showed the limited effectiveness of such interventions. We propose a paradigm-shift in online moderation by moving towards a personalized and user-centered approach. Our multidisciplinary vision combines state-of-the-art theories and practices in diverse fields such as computer science, sociology and psychology, to design personalized moderation interventions (PMIs). In outlining the path leading to the next-generation of moderation interventions, we also discuss the most prominent challenges introduced by such a disruptive change.
The integration of AI into various aspects of our lives has significantly reshaped how we access information, products, and services. AI-driven personalization, a key feature of many platforms, aims to enhance user experiences by tailoring content to individual preferences. Advancement of AI personalization has given rise to the filter bubble and echo-chamber phenomena. Filter bubbles expose consumers to content that reinforces their existing beliefs and preferences, creating a paradox. This chapter explores the multifaceted implications of AI-personalization paradox on digital consumer behavior in the context of the filter bubble era investigating how AI algorithms shape the content users encounter, impact of algorithms on information diversity, and consequences for consumer decision-making. The chapter concludes speculating on future of personalization, emphasizing the need to balance customization with information diversity, encouraging critical thinking about AI ethics among consumers.
This research paper analysis Spotify data using Python to investigate the characteristics contributing to song popularity. The objectives are to assess the popularity index, identify key attributes of popular songs, and develop a model for predicting song popularity based on current characteristics. The analysis involves data cleaning, exploratory data analysis, and visualization using Python libraries. With over 381 million monthly active users, Spotify provides a rich dataset for understanding music listening habits. Previous studies have explored Spotify's technologies and popularity, enhancing understanding of its protocols and user behavior. This research paper aims to uncover patterns and relationships within the data by applying statistical and machine-learning techniques. The findings will inform actionable recommendations and contribute to a better understanding of music consumption patterns and preferences.
GİRİŞ Bir müzik dinleme uygulaması olan Spotify'ın web sitesine girdiğinizde sizi büyük fontlarla şu cümle karşılar: "Dinlemek her şeydir". Milyonlarca mü-ziği, podcasti (dijital ses yayını) telefon, bilgisayar, tablet, akıllı saatler vb. bir-çok cihazda dinleme imkânı sunan Spotify, direksiyon başında, spor yaparken, eğlenirken veya dinlenirken doğru müzik ya da podcastin veya önerilerin her zaman kullanıcının "parmaklarının ucunda" olduğunu belirterek kendini ta-nımlamaktadır (spotify.com, et: 2022). İnternet teknolojisinin Web 3.0 seviye-sine gelmesiyle dinamik, kişisel, kreatif, akıllı uygulamaların olduğu bir dün-yaya da giriş yapılmıştır. Günümüzde ise simbiyotik web olarak bilinen, maki-nelerin ve insanların etkileşimde olacağı, makinelerin insanlara en kaliteli ve anlamlı bilgiyi ulaştıracağı bir teknoloji sistemi olan Web 4.0 konuşulmakta-dır. (Aghaei vd., 2012) Bu teknolojinin yapay zekâ ve makine öğrenmesinin katkısıyla insan hayatında önemli değişimler yaratacağı düşünülmektedir. Web 4.0 teknolojisinin gündelik hayattaki farklı örnekleri her geçen gün daha fazla görünür hâle gelmektedir. Akıllı telefonlar, mobil uygulamalar ve kişisel-leştirilmiş profilleri yoğun bir şekilde deneyimleyen internet kullanıcıları; giyilebilir teknolojiler, nesnelerin interneti teknolojisiyle cihazlar arası ileti-şimin insanlara olan hizmeti gibi birçok gelişmeyi merakla takip etmektedir. Yapay zekâyı kullanmaya başlayan birçok uygulama gibi Spotify da kullanıcıla-rının uygulamadaki tutum ve hareketlerinden kişiselleştirilmiş bir eğlence modeli çıkarmakta ve bunları yapay zekâ teknolojisiyle gerçekleştirmektedir. Mobil uygulamalar, teknolojinin gelişmesiyle birlikte hayatımıza giren, ile-tişim, haberleşme, eğlence, alışveriş, bilgi edinme gibi ihtiyaçları mobil yani ha-reket hâlindeyken telefon, tablet, taşınabilir akıllı cihazlar vasıtasıyla gidermeye
Artificial intelligence (A.I.) increasingly suffuses everyday life. However, people are frequently reluctant to interact with A.I. systems. This challenges both the deployment of beneficial A.I. technology and the development of deep learning systems that depend on humans for oversight, direction, and regulation. Nine studies (N = 3,300) demonstrate that social-cognitive processes guide human interactions across a diverse range of real-world A.I. systems. Across studies, perceived warmth and competence emerge prominently in participants' impressions of A.I. systems. Judgments of warmth and competence systematically depend on human-A.I. interdependence and autonomy. In particular, participants perceive systems that optimize interests aligned with human interests as warmer and systems that operate independently from human direction as more competent. Finally, a prisoner's dilemma game shows that warmth and competence judgments predict participants' willingness to cooperate with a deep-learning system. These results underscore the generality of intent detection to perceptions of a broad array of algorithmic actors.
Background
Empirical support for the notion that music listening is beneficial for stress recovery is inconclusive, potentially due to the methodological diversity with which the effects of music on stress recovery have been investigated. Little is presently known about which recovery activities are chosen by individuals for the purpose of stress recovery, and whether audio feature commonalities exist between different songs that are selected by individuals for the purpose of stress recovery. The current pre-registered study investigated whether audio feature commonalities can be extracted from self-selected songs for the purpose of stress recovery. Furthermore, the present study exploratorily examined the relationship between audio features and participants’ desired recovery-related emotions while listening and after listening to self-selected music.
Methods
Participants (N = 470) completed an online survey in which they described what music they would listen to unwind from a hypothetical stressful event. Data analysis was conducted using a split-sample procedure. A k-medoid cluster analysis was conducted to identify audio feature commonalities between self-selected songs. Multiple regression analyses were conducted to examine the relationship between audio features and desired recovery emotions.
Results
Participants valued music listening as a recovery activity to a similar extent as watching TV, sleeping, or talking to a significant other. Cluster analyses revealed that self-selected songs for the purpose of stress recovery can be grouped into two distinct categories. The two categories of songs shared similarities in key, loudness, speechiness, acousticness, instrumentalness, liveness, musical valence, tempo, duration, and time signature, and were distinguished by danceability, energy, and mode. No audio features were significantly associated with participants’ desired recovery emotions.
Conclusions
Although a comprehensive portrait of the relationship between audio features and stress recovery still warrants further research, the present study provides a starting point for future enquiries into the nuanced effects of musical audio features on stress recovery.
In this modern era, many people from various backgrounds listen to music through digital album applications. This study was made to analyze one of the features of digital music applications that have many listeners in various parts of Indonesia. This study aims to describe the meaning of the denotation "Mendengarkan Secara Offline" on the feature in the Spotify application. In this study, a descriptive qualitative approach was used to reveal the meaning of the denotation "Mendengarkan Secara Offline". The technique of data collection was done by using sentence analysis. This study concludes that the denotation meaning of "Mendengarkan Secara Offline" feature in the Spotify application means that music in online music applications can be listened while being offline. It can be listened without having an internet connection. The music could be saved using any storage in the phone's memory. "Mendengarkan Secara Offline" feature allows the users listen to music easily and it offers convenience to users listen to music freely.
An accurate personality model is crucial to many research fields. Most personality models have been constructed using linear factor analysis (LFA). In this paper, we investigate if an effective deep learning tool for factor extraction, the Variational Autoencoder (VAE), can be applied to explore the factor structure of a set of personality variables. To compare VAE with LFA, we applied VAE to an International Personality Item Pool (IPIP) Big 5 dataset and an IPIP HEXACO (Humility-Honesty, Emotionality, Extroversion, Agreeableness, Conscientiousness, Openness) dataset. We found that LFA tends to break factors into ever smaller, yet still significant fractions, when the number of assumed latent factors increases, leading to the need to organize personality variables at the factor level and then the facet level. On the other hand, the factor structure returned by VAE is very stable and VAE only adds noise-like factors after significant factors are found as the number of assumed latent factors increases. VAE reported more stable factors by elevating some facets in the HEXACO scale to the factor level. Since this is a data-driven process that exhausts all stable and significant factors that can be found, it is not necessary to further conduct facet level analysis and it is anticipated that VAE will have broad applications in exploratory factor analysis in personality research.
Federated learning (FL) is an effective mechanism for data privacy in recommender systems by running machine learning model training on-device. While prior FL optimizations tackled the data and system heterogeneity challenges faced by FL, they assume the two are independent of each other. This fundamental assumption is not reflective of real-world, large-scale recommender systems -- data and system heterogeneity are tightly intertwined. This paper takes a data-driven approach to show the inter-dependence of data and system heterogeneity in real-world data and quantifies its impact on the overall model quality and fairness. We design a framework, RF^2, to model the inter-dependence and evaluate its impact on state-of-the-art model optimization techniques for federated recommendation tasks. We demonstrate that the impact on fairness can be severe under realistic heterogeneity scenarios, by up to 15.8--41x compared to a simple setup assumed in most (if not all) prior work. It means when realistic system-induced data heterogeneity is not properly modeled, the fairness impact of an optimization can be downplayed by up to 41x. The result shows that modeling realistic system-induced data heterogeneity is essential to achieving fair federated recommendation learning. We plan to open-source RF^2 to enable future design and evaluation of FL innovations.
The exponential growth of online services and user data changed how we interact with various services, and how we explore and select new products. Hence, there is a growing need for methods to recommend the appropriate items for each user. In the case of music, it is more important to recommend the right items at the right moment. It has been well documented that the context, i.e. the listening situation of the users, strongly influences their listening preferences. Hence, there has been an increasing attention towards developing recommendation systems. State-of-the-art approaches are sequence-based models aiming at predicting the tracks in the next session using available contextual information. However, these approaches lack interpretability and serve as a hit-or-miss with no room for user involvement. Additionally, few previous approaches focused on studying how the audio content relates to these situational influences, and even to a less extent making use of the audio content in providing contextual recommendations. Hence, these approaches suffer from both lack of interpretability.In this dissertation, we study the potential of using the audio content primarily to disambiguate the listening situations, providing a pathway for interpretable recommendations based on the situation.First, we study the potential listening situations that influence/change the listening preferences of the users. We developed a semi-automated approach to link between the listened tracks and the listening situation using playlist titles as a proxy. Through this approach, we were able to collect datasets of music tracks labelled with their situational use. We proceeded with studying the use of music auto-taggers to identify potential listening situations using the audio content. These studies led to the conclusion that the situational use of a track is highly user-dependent. Hence, we proceeded with extending the music-autotaggers to a user-aware model to make personalized predictions. Our studies showed that including the user in the loop significantly improves the performance of predicting the situations. This user-aware music auto-tagger enabled us to tag a given track through the audio content with potential situational use, according to a given user by leveraging their listening history.Finally, to successfully employ this approach for a recommendation task, we needed a different method to predict the potential current situations of a given user. To this end, we developed a model to predict the situation given the data transmitted from the user's device to the service, and the demographic information of the given user. Our evaluations show that the models can successfully learn to discriminate the potential situations and rank them accordingly. By combining the two model; the auto-tagger and situation predictor, we developed a framework to generate situational sessions in real-time and propose them to the user. This framework provides an alternative pathway to recommending situational sessions, aside from the primary sequential recommendation system deployed by the service, which is both interpretable and addressing the cold-start problem in terms of recommending tracks based on their content.
This chapter investigates changes in land use intensity in a crop-livestock farming system on Lemnos Island through the combination of land use/land cover (LULC) types extracted from black and white (B&W) aerial images, statistics, and qualitative data. Combining quantitative and qualitative data, different insights of land use and landscape changes are assessed. The steps are: first, the timeline of historical changes was compiled; then, remote sensing was used to assess land cover changes and conversions that represent changes in intensity, and this information was complemented by participatory mapping with local farmers. Land use trajectories revealed that extensification is the basic trend from 1960 to 2002. Intensification coexists in at the same time, as grasslands convert to crops. There is a distinct pattern between periods: extensification seems to be the main process of change from 1960 to 1980 affecting mainly the hilly uplands as more remote and marginal areas are being converted from crops to grasslands or are abandoned, whereas intensification is the main trend for the next 20 years mainly in the lowlands as modernized agriculture (irrigated fields, land aggregation, machinery use) replaces more extensive land uses and traditional landscape elements such as tree hedges. The role of complementarity is very important. Conclusively, this case study shows that in some farming systems land use intensity changes cannot be represented through simple dichotomist differentiations.KeywordsLand use intensityAerial panchromatic imagesImage segmentationParticipatory mappingCrop-livestock farming system
ResearchGate has not been able to resolve any references for this publication.