No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Improving Effectiveness by Reducing Overconfidence in Large Catalogue Sequential Recommendation with gBCE loss
Abstract
A large catalogue size is one of the central challenges in training recommendation models: a large number of items makes them memory and computationally inefficient to compute scores for all items during training, forcing these models to deploy negative sampling. However, negative sampling increases the proportion of positive interactions in the training data, and therefore, models trained with negative sampling tend to overestimate the probabilities of positive interactions – a phenomenon we call overconfidence . While the absolute values of the predicted scores/probabilities are not important for the ranking of retrieved recommendations, overconfident models may fail to estimate nuanced differences in the top-ranked items, resulting in degraded performance. In this paper, we show that overconfidence explains why the popular SASRec model underperforms when compared to BERT4Rec. This is contrary to the BERT4Rec authors’ explanation that the difference in performance is due to the bi-directional attention mechanism. To mitigate overconfidence, we propose a novel Generalised Binary Cross-Entropy Loss function (gBCE) and theoretically prove that it can mitigate overconfidence. We further propose the gSASRec model, an improvement over SASRec that deploys an increased number of negatives and the gBCE loss. Through detailed experiments on three datasets, we show that gSASRec does not exhibit the overconfidence problem. As a result, gSASRec can outperform BERT4Rec (e.g. +9.47% NDCG on the MovieLens-1M dataset), while requiring less training time (e.g. -73% training time on MovieLens-1M). Moreover, in contrast to BERT4Rec, gSASRec is suitable for large datasets that contain more than 1 million items. Finally, we show how addressing overconfidence can improve model calibration – the ability of a model to predict actual interaction probabilities accurately. By applying gBCE to the SASRec model on MovieLens-1M dataset, we reduce the models’ expected calibration error by 98.9% (from 0.966 to 0.01).
ResearchGate has not been able to resolve any citations for this publication.
A large catalogue size is one of the central challenges in training recommendation models: a large number of items makes them memory and computationally inefficient to compute scores for all items during training, forcing these models to deploy negative sampling. However, negative sampling increases the proportion of positive interactions in the training data, and therefore models trained with negative sampling tend to overestimate the probabilities of positive interactions-a phenomenon we call overconfidence. While the absolute values of the predicted scores/probabilities are not important for the ranking of retrieved recommendations, overconfident models may fail to estimate nuanced differences in the top-ranked items, resulting in degraded performance. In this paper, we show that over-confidence explains why the popular SASRec model underperforms when compared to BERT4Rec. This is contrary to the BERT4Rec authors' explanation that the difference in performance is due to the bi-directional attention mechanism. To mitigate overconfidence, we propose a novel Generalised Binary Cross-Entropy Loss function (gBCE) and theoretically prove that it can mitigate overconfidence. We further propose the gSASRec model, an improvement over SAS-Rec that deploys an increased number of negatives and the gBCE loss. We show through detailed experiments on three datasets that gSASRec does not exhibit the overconfidence problem. As a result, gSASRec can outperform BERT4Rec (e.g. +9.47% NDCG on the MovieLens-1M dataset) while requiring less training time (e.g.-73% training time on MovieLens-1M). Moreover, in contrast to BERT4Rec, gSASRec is suitable for large datasets that contain more than 1 million items.
Sequential recommendation is an important recommendation task that aims to predict the next item in a sequence. Recently, adaptations of language models, particularly Transformer-based models such as SASRec and BERT4Rec, have achieved state-of-the-art results in sequential recommendation. In these models, item ids replace tokens in the original language models. However, this approach has limitations. First, the vocabulary of item ids may be many times larger than in language models. Second, the classical Top-K recommendation approach used by these models may not be optimal for complex recommendation objectives, including auxiliary objectives such as diversity, coverage or coherence. Recent progress in generative language models inspires us to revisit gener-ative approaches to address these challenges. This paper presents the GPTRec sequential recommendation model, which is based on the GPT-2 architecture. GPTRec can address large vocabulary issues by splitting item ids into sub-id tokens using a novel SVD Tokeni-sation algorithm based on quantised item embeddings from a SVD decomposition of the user-item interaction matrix. The paper also presents a novel Next-K recommendation strategy, which generates recommendations item-by-item, considering already recommended items. The Next-K strategy can be used for producing complex interdependent recommendation lists. We experiment with GPTRec on the MovieLens-1M dataset and show that using sub-item tokeni-sation GPTRec can match the quality of SASRec while reducing the embedding table by 40%. We also show that the recommendations generated by GPTRec on MovieLens-1M using the Next-K recommendation strategy match the quality of SASRec in terms of NDCG@10, meaning that the model can serve as a strong starting point for future research.
We address the problem of user intent prediction from clickstream data of an e-commerce website via two conceptually different approaches: a hand-crafted feature-based classification and a deep learning-based classification. In both approaches, we deliberately coarse-grain a new clickstream proprietary dataset to produce symbolic trajectories with minimal information. Then, we tackle the problem of trajectory classification of arbitrary length and ultimately, early prediction of limited-length trajectories, both for balanced and unbalanced datasets. Our analysis shows that k-gram statistics with visibility graph motifs produce fast and accurate classifications, highlighting that purchase prediction is reliable even for extremely short observation windows. In the deep learning case, we benchmarked previous state-of-the-art (SOTA) models on the new dataset, and improved classification accuracy over SOTA performances with our proposed LSTM architecture. We conclude with an in-depth error analysis and a careful evaluation of the pros and cons of the two approaches when applied to realistic industry use cases.
Popularity is often included in experimental evaluation to provide a reference performance for a recommendation task. To understand how popularity baseline is defined and evaluated, we sample 12 papers from top-tier conferences including KDD, WWW, SIGIR, and RecSys, and 6 open source toolkits. We note that the widely adopted MostPop baseline simply ranks items based on the number of interactions in the training data. We argue that the current evaluation of popularity (i) does not reflect the popular items at the time when a user interacts with the system, and (ii) may recommend items released after a user's last interaction with the system. On the widely used MovieLens dataset, we show that the performance of popularity could be significantly improved by 70% or more, if we consider the popular items at the time point when a user interacts with the system. We further show that, on MovieLens dataset, the users having lower tendencies on movies tend to follow the crowd and rate more popular movies. Movie lovers who rate a large number of movies, rate movies based on their own preferences and interests. Through this study, we call for a re-visit of the popularity baseline in recommender system to better reflect its effectiveness.
In this paper, we present a comprehensive review of the imbalance problems in object detection. To analyze the problems in a systematic manner, we introduce a problem-based taxonomy. Following this taxonomy, we discuss each problem in depth and present a unifying yet critical perspective on the solutions in the literature. In addition, we identify major open issues regarding the existing imbalance problems as well as imbalance problems that have not been discussed before. Moreover, in order to keep our review up to date, we provide an accompanying webpage which catalogs papers addressing imbalance problems, according to our problem-based taxonomy. Researchers can track newer studies on this webpage available at: https://github.com/kemaloksuz/ObjectDetectionImbalance.
In the evaluation of magnetic field data, edge enhancement and detection
techniques are important treatments for the interpretation of geological structures. In
general geological sense, contiguity of deep and shallow magnetic sources leads to weak
and intense anomalies that complicates the interpretation to disclose adjacent anomalous
sources. Many of the existing filters for edge detection in magnetics mostly have the disadvantage
that they require a reduction to pole transformation as the pre-process of the
data or they cannot balance weak and intense anomalies and therefore fail in detecting
edges of deep and shallow sources simultaneously. This study presents an improved edge
detection filter LAS (logistic function of the analytical signal), based on the generalised
logistic function configured by the ratio of derivatives of the analytical signal. This novel
approach has the capability of reducing the dependence on the direction of the magnetization
and also balancing anomalies of sources at different levels of depth. The feasibility of
the method is examined on both theoretical and real data cases comparatively with some
other methods that utilize the analytical signal in their basis. In comparison, the results
demonstrate that the LAS method provides more accurate estimation of edge localization.
Convolutional Neural Networks (CNNs) have been recently introduced in the domain of session-based next item recommendation. An ordered collection of past items the user has interacted with in a session (or sequence) are embedded into a 2-dimensional latent matrix, and treated as an image. The convolution and pooling operations are then applied to the mapped item embeddings. In this paper, we first examine the typical session-based CNN recommender and show that both the generative model and network architecture are suboptimal when modeling long-range dependencies in the item sequence. To address the issues, we introduce a simple, but very effective generative model that is capable of learning high-level representation from both short- and long-range item dependencies. The network architecture of the proposed model is formed of a stack of holed convolutional layers, which can efficiently increase the receptive fields without relying on the pooling operation. Another contribution is the effective use of residual block structure in recommender systems, which can ease the optimization for much deeper networks. The proposed generative model attains state-of-the-art accuracy with less training time in the next item recommendation task. It accordingly can be used as a powerful recommendation baseline to beat in future, especially when there are long sequences of user feedback.
RNNs have been shown to be excellent models for sequential data and in particular for data that is generated by users in an session-based manner. The use of RNNs provides impressive performance benefits over classical methods in session-based recommendations. In this work we introduce novel ranking loss functions tailored to RNNs in the recommendation setting. The improved performance of these losses over alternatives, along with further tricks and refinements described in this work, allow for an overall improvement of up to 35% in terms of MRR and [email protected] over previous session-based RNN solutions and up to 53% over classical collaborative filtering approaches. Unlike data augmentation-based improvements, our method does not increase training times significantly. We further demonstrate the performance gain of the RNN over baselines in an online A/B test.
Recommender systems have been widely used in various domains including movies, news, music with an aim to provide the most relevant proposals to users from a variety of available options. Recommender systems are designed using techniques from many fields, some of which are: machine learning, information retrieval, data mining, linear algebra and artificial intelligence. Though in-memory nearest-neighbor computation is a typical approach for collaborative filtering due to its high recommendation accuracy; its performance on scalability is still poor given a huge user and item base and availability of only few ratings (i.e., data sparsity) in archetypal merchandising applications. In order to alleviate scalability and sparsity issues in recommender systems, several model-based approaches were proposed in the past. However, if research in recommender system is to achieve its potential, there is a need to understand the prominent techniques used directly to build recommender systems or for preprocessing recommender datasets, along with its strengths and weaknesses. In this work, we present an overview of some of the prominent traditional as well as advanced techniques that can effectively handle data dimensionality and data sparsity. The focus of this survey is to present an overview of the applicability of some advanced techniques, particularly clustering, biclustering, matrix factorization, graph-theoretic, and fuzzy techniques in recommender systems. In addition, it highlights the applicability and recent research works done using each technique.
Top-N sequential recommendation models each user as a sequence of items interacted in the past and aims to predict top-N ranked items that a user will likely interact in a »near future». The order of interaction implies that sequential patterns play an important role where more recent items in a sequence have a larger impact on the next item. In this paper, we propose a Convolutional Sequence Embedding Recommendation Model »Caser» as a solution to address this requirement. The idea is to embed a sequence of recent items into an »image» in the time and latent spaces and learn sequential patterns as local features of the image using convolutional filters. This approach provides a unified and flexible network structure for capturing both general preferences and sequential patterns. The experiments on public data sets demonstrated that Caser consistently outperforms state-of-the-art sequential recommendation methods on a variety of common evaluation metrics.
RNNs have been shown to be excellent models for sequential data and in particular for session-based user behavior. The use of RNNs provides impressive performance benefits over classical methods in session-based recommendations. In this work we introduce a novel ranking loss function tailored for RNNs in recommendation settings. The better performance of such loss over alternatives, along with further tricks and improvements described in this work, allow to achieve an overall improvement of up to 35% in terms of MRR and Recall@20 over previous session-based RNN solutions and up to 51% over classical collaborative filtering approaches. Unlike data augmentation-based improvements, our method does not increase training times significantly.
Transformer-based Cross-Encoders achieve state-of-the-art effectivness in text retrieval. However, Cross-Encoders based on large transformer models (such as BERT or T5) are computationally expensive and allow for scoring only a small number of documents within a reasonably small latency window. However, keeping search latencies low is important for user satisfaction and energy usage. In this paper, we show that weaker shallow transformer models (i.e. transformers with a limited number of layers) actually perform better than full-scale models when constrained to these practical low-latency settings, since they can estimate the relevance of more documents in the same time budget. We further show that shallow transformers may benefit from the generalised Binary Cross-Entropy (gBCE) training scheme, which has recently demonstrated success for recommendation tasks. Our experiments with TREC Deep Learning passage ranking querysets demonstrate significant improvements in shallow and full-scale models in low-latency scenarios. For example, when the latency limit is 25 ms per query, MonoBERT-Large (a cross-encoder based on a full-scale BERT model) is only able to achieve NDCG@10 of 0.431 on TREC DL 2019, while TinyBERT-gBCE (a cross-encoder based on TinyBERT trained with gBCE) reaches NDCG@10 of 0.652, a +51% gain over MonoBERT-Large. We also show that shallow Cross-Encoders are effective even when used without a GPU (e.g., with CPU inference, NDCG@10 decreases only by 3% compared to GPU inference with 50 ms latency), which makes Cross-Encoders practical to run even without specialised hardware acceleration.
Sequential Recommendation is a popular recommendation task that uses the order of user-item interaction to model evolving users' interests and sequential patterns in their behaviour. Current state-of-the-art Transformer-based models for sequential recommendation, such as BERT4Rec and SASRec, generate sequence embeddings and compute scores for catalogue items, but the increasing catalogue size makes training these models costly. The Joint Product Quan-tisation (JPQ) method, originally proposed for passage retrieval, markedly reduces the size of the retrieval index with minimal effect on model effectiveness, by replacing passage embeddings with a limited number of shared sub-embeddings. This paper introduces RecJPQ, a novel adaptation of JPQ for sequential recommendations, which takes the place of item embeddings tensor and replaces item embeddings with a concatenation of a limited number of shared sub-embeddings and, therefore, limits the number of learnable model parameters. The main idea of RecJPQ is to split items into sub-item entities before training the main recommendation model, which is inspired by splitting words into tokens and training tokenisers in language models. We apply RecJPQ to SASRec, BERT4Rec, and GRU4rec models on three large-scale sequential datasets. Our results showed that RecJPQ could notably reduce the model size (e.g., 48× reduction for the Gowalla dataset with no effectiveness degradation). RecJPQ can also improve model performance through a regularisation effect (e.g. +0.96% NDCG@10 improvement on the Booking.com dataset). Overall, RecJPQ allows the training of state-of-the-art transformer recommenders in industrial applications, where datasets with millions of items are common.
The learning objective plays a fundamental role to build a recommender system. Most methods routinely adopt either pointwise ( e.g., binary cross-entropy) or pairwise ( e.g., BPR) loss to train the model parameters, while rarely pay attention to softmax loss, which assumes the probabilities of all classes sum up to 1, due to its computational complexity when scaling up to large datasets or intractability for streaming data where the complete item space is not always available. The sampled softmax (SSM) loss emerges as an efficient substitute for softmax loss. Its special case, InfoNCE loss, has been widely used in self-supervised learning and exhibited remarkable performance for contrastive learning. Nonetheless, limited recommendation work uses the SSM loss as the learning objective. Worse still, none of them explores its properties thoroughly and answers “Does SSM loss suit for item recommendation?” and “What are the conceptual advantages of SSM loss, as compared with the prevalent losses?”, to the best of our knowledge.
In this work, we aim to offer a better understanding of SSM for item recommendation. Specifically, we first theoretically reveal three model-agnostic advantages: (1) mitigating popularity bias, which is beneficial to long-tail recommendation; (2) mining hard negative samples, which offers informative gradients to optimize model parameters; and (3) maximizing the ranking metric, which facilitates top- K performance. However, based on our empirical studies, we recognize that the default choice of cosine similarity function in SSM limits its ability in learning the magnitudes of representation vectors. As such, the combinations of SSM with the models that also fall short in adjusting magnitudes ( e.g., matrix factorization) may result in poor representations. One step further, we provide mathematical proof that message passing schemes in graph convolution networks can adjust representation magnitude according to node degree, which naturally compensates for the shortcoming of SSM. Extensive experiments on four benchmark datasets justify our analyses, demonstrating the superiority of SSM for item recommendation. Our implementations are available in both TensorFlow and PyTorch.
Many modern sequential recommender systems use deep neural networks, which can effectively estimate the relevance of items, but require a lot of time to train. Slow training increases the costs of training, hinders product development timescales and prevents the model from being regularly updated to adapt to changing user preferences. The training of such sequential models involves appropriately sampling past user interactions to create a realistic training objective. The existing training objectives have limitations. For instance, next item prediction never uses the beginning of the sequence as a learning target, thereby potentially discarding valuable data. On the other hand, the item masking used by the state-of-the-art BERT4Rec recommender model is only weakly related to the goal of the sequential recommendation; therefore, it requires much more time to obtain an effective model. Hence, we propose a novel Recency-based Sampling of Sequences (RSS) training objective (which is parameterized by a choice of recency importance function) that addresses both limitations. We apply our method to various recent and state-of-the-art model architectures – such as GRU4Rec, Caser, and SASRec. We show that the models enhanced with our method can achieve performances exceeding or very close to the effective BERT4Rec, but with much less training time. For example, on the MovieLens-20M dataset, RSS applied to the SASRec model can result in a 60% improvement in NDCG over a vanilla SASRec, and a 16% improvement over a fully-trained BERT4Rec model, despite taking 93% less training time than BERT4Rec. We also experiment with two families of recency importance functions and show that they perform similarly. We further empirically demonstrate that RSS-enhanced SASRec successfully learns to distinguish differences between recent and older interactions – a property that the original SASRec model does not exhibit. Overall, we show that RSS is a viable (and frequently better) alternative to the existing training objectives, which is both effective and efficient for training sequential recommender model when the computational resources for training are limited.
The task of item recommendation is to select the best items for a user from a large catalogue of items. Item recommenders are commonly trained from implicit feedback which consists of past actions that are positive only. Core challenges of item recommendation are (1) how to formulate a training objective from implicit feedback and (2) how to efficiently train models over a large item catalogue. This chapter formulates the item recommendation problem and points out its unique characteristics. Then different training objectives are discussed. The main body deals with learning algorithms and presents sampling based algorithms for general recommenders and more efficient algorithms for dot product models. Finally, the application of item recommenders for retrieval tasks is discussed.
Recommender systems personalize content by recommending items to users. Item recommendation algorithms are evaluated by metrics that compare the positions of truly relevant items among the recommended items. To speed up the computation of metrics, recent work often uses sampled metrics where only a smaller set of random items and the relevant items are ranked. This paper investigates such sampled metrics in more detail and shows that they are inconsistent with their exact counterpart, in the sense that they do not persist relative statements, for example, recommender A is better than B , not even in expectation. Moreover, the smaller the sample size, the less difference there is between metrics, and for very small sample size, all metrics collapse to the AUC metric. We show that it is possible to improve the quality of the sampled metrics by applying a correction, obtained by minimizing different criteria. We conclude with an empirical evaluation of the naive sampled metrics and their corrected variants. To summarize, our work suggests that sampling should be avoided for metric calculation, however if an experimental study needs to sample, the proposed corrections can improve the quality of the estimate.
Modeling users' dynamic preferences from their historical behaviors is challenging and crucial for recommendation systems. Previous methods employ sequential neural networks to encode users' historical interactions from left to right into hidden representations for making recommendations. Despite their effectiveness, we argue that such left-to-right unidirectional models are sub-optimal due to the limitations including: \begin enumerate* [label=series\itshape\alph*\upshape)] \item unidirectional architectures restrict the power of hidden representation in users' behavior sequences; \item they often assume a rigidly ordered sequence which is not always practical. \end enumerate* To address these limitations, we proposed a sequential recommendation model called BERT4Rec, which employs the deep bidirectional self-attention to model user behavior sequences. To avoid the information leakage and efficiently train the bidirectional model, we adopt the Cloze objective to sequential recommendation, predicting the random masked items in the sequence by jointly conditioning on their left and right context. In this way, we learn a bidirectional representation model to make recommendations by allowing each item in user historical behaviors to fuse information from both left and right sides. Extensive experiments on four benchmark datasets show that our model outperforms various state-of-the-art sequential models consistently.
Many websites offer promotions in terms of bundled items that can be purchased together, usually at a discounted rate. 'Bundling' may be a means of increasing sales revenue, but may also be a means for content creators to expose users to new items that they may not have considered in isolation. In this paper, we seek to understand the semantics of what constitutes a 'good' bundle, in order to recommend existing bundles to users on the basis of their constituent products, as well the more difficult task of generating new bundles that are personalized to a user. To do so we collect a new dataset from the Steam video game distribution platform, which is unique in that it contains both 'traditional' recommendation data (rating and purchase histories between users and items), as well as bundle purchase information. We assess issues such as bundle size and item compatibility, and show that these features, when combined with traditional matrix factorization techniques, can lead to highly effective bundle recommendation and generation.
Confidence calibration -- the problem of predicting probability estimates representative of the true correctness likelihood -- is important for classification models in many applications. We discover that modern neural networks, unlike those from a decade ago, are poorly calibrated. Through extensive experiments, we observe that depth, width, weight decay, and Batch Normalization are important factors influencing calibration. We evaluate the performance of various post-processing calibration methods on state-of-the-art architectures with image and document classification datasets. Our analysis and experiments not only offer insights into neural network learning, but also provide a simple and straightforward recipe for practical settings: on most datasets, temperature scaling -- a single-parameter variant of Platt Scaling -- is surprisingly effective at calibrating predictions.
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.0 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.
In recent years, deep neural networks have yielded immense success on speech recognition, computer vision and natural language processing. However, the exploration of deep neural networks on recommender systems has received relatively less scrutiny. In this work, we strive to develop techniques based on neural networks to tackle the key problem in recommendation --- collaborative filtering --- on the basis of implicit feedback.
Although some recent work has employed deep learning for recommendation, they primarily used it to model auxiliary information, such as textual descriptions of items and acoustic features of musics. When it comes to model the key factor in collaborative filtering --- the interaction between user and item features, they still resorted to matrix factorization and applied an inner product on the latent features of users and items.
By replacing the inner product with a neural architecture that can learn an arbitrary function from data, we present a general framework named NCF, short for Neural network-based Collaborative Filtering. NCF is generic and can express and generalize matrix factorization under its framework. To supercharge NCF modelling with non-linearities, we propose to leverage a multi-layer perceptron to learn the user-item interaction function. Extensive experiments on two real-world datasets show significant improvements of our proposed NCF framework over the state-of-the-art methods. Empirical evidence shows that using deeper layers of neural networks offers better recommendation performance.
State-of-the-art item recommendation algorithms, which apply Factorization Machines (FM) as a scoring function and pairwise ranking loss as a trainer (PRFM for short), have been recently investigated for the implicit feedback based context-aware recommendation problem (IFCAR). However, good recommenders particularly emphasize on the accuracy near the top of the ranked list, and typical pairwise loss functions might not match well with such a requirement. In this paper, we demonstrate, both theoretically and empirically, PRFM models usually lead to non-optimal item recommendation results due to such a mismatch. Inspired by the success of LambdaRank, we introduce Lambda Factorization Machines (LambdaFM), which is particularly intended for optimizing ranking performance for IFCAR. We also point out that the original lambda function suffers from the issue of expensive computational complexity in such settings due to a large amount of unobserved feedback. Hence, instead of directly adopting the original lambda strategy, we create three effective lambda surrogates by conducting a theoretical analysis for lambda from the top-N optimization perspective. Further, we prove that the proposed lambda surrogates are generic and applicable to a large set of pairwise ranking loss functions. Experimental results demonstrate LambdaFM significantly outperforms state-of-the-art algorithms on three real-world datasets in terms of four standard ranking measures.
YouTube represents one of the largest scale and most sophisticated industrial recommendation systems in existence. In this paper, we describe the system at a high level and focus on the dramatic performance improvements brought by deep learning. The paper is split according to the classic two-stage information retrieval dichotomy: first, we detail a deep candidate generation model and then describe a separate deep ranking model. We also provide practical lessons and insights derived from designing, iterating and maintaining a massive recommendation system with enormous user-facing impact.
The MovieLens datasets are widely used in education, research, and industry. They are downloaded hundreds of thousands of times each year, reflecting their use in popular press programming books, traditional and online courses, and software. These datasets are a product of member activity in the MovieLens movie recommendation system, an active research platform that has hosted many experiments since its launch in 1997. This article documents the history of MovieLens and the MovieLens datasets. We include a discussion of lessons learned from running a long-standing, live research platform from the perspective of a research organization. We document best practices and limitations of using the MovieLens datasets in new research.