BPR: Bayesian Personalized Ranking from Implicit Feedback
By Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, Lars Schmidt-Thieme. 2009
6,726 citations recorded by the source corpus.
Item recommendation is the task of predicting a personalized ranking on a set of items (e.g. websites, movies, products). In this paper, we investigate the most common scenario with implicit feedback (e.g. clicks, purchases). There are many methods for item recommendation from implicit feedback like matrix factorization (MF) or adaptive k-nearest-neighbor (kNN). Even though these methods are designed for the item prediction task of personalized ranking, none of them is directly optimized for ranking. In this paper we present a generic optimization criterion BPR-Opt for personalized ranking that is the maximum posterior estimator derived from a Bayesian analysis of the problem. We also provide a generic learning algorithm for optimizing models with respect to BPR-Opt. The learning method is based on stochastic gradient descent with bootstrap sampling. We show how to apply our method to two state-of-the-art recommender models: matrix factorization and adaptive kNN. Our experiments indicate that for the task of personalized ranking our optimization method outperforms the standard learning techniques for MF and kNN. The results show the importance of optimizing models for the right criterion.
Subfields: Artificial Intelligence, Information Systems, Recommender Systems
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Xiangnan He
- Mohit Bansal
- Tat-seng Chua
- Jianfeng Gao
- Yoshua Bengio
- Percy Liang
- Quoc V. Le
- Dacheng Tao
In Information Systems
- Xiangnan He
- Yongfeng Zhang
- Tat-seng Chua
- Xu Chen
- Xiang Wang
- Julian McAuley
- Xing Xie
- Charu C. Aggarwal
In Recommender Systems
Citations in the corpus
- Multi-step Retriever-Reader Interaction for Scalable Open-domain Question Answering (2019, methodology): “We follow the Bayesian Personalized Ranking approach of Rendle et al. (2009) and maximize log ( σ ( q (cid:62) p ∗ − q (cid:62) p )) .”
- Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering (2016, methodology): “Bayesian Personalized Ranking (BPR) is a pairwise ranking optimization framework which adopts Stochastic Gradient Ascent to optimize the regularized corpus likelihood [30].”
- Variational Autoencoders for Collaborative Filtering (2018, methodology): “We also experiment with Bayesian personalized ranking (bpr) [36].”
- VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback (2015, background): “proposed to uncover the most relevant latent dimensions in both explicit and implicit feedback settings (Rendle et al. 2009; Bell, Koren, and Volinsky 2007; Pan et al. 2008; Hu, Koren, and Volinsky 2008).”
- Recurrent Latent Variable Networks for Session-Based Recommendation (2017, methodology): “Since BPR-MF is designed for processing single item feature vectors, as oppposed to sequences , to apply it in the context of session-based recommendation we average the feature vectors of the items occurring thus far in a given session.”
- Neural Collaborative Filtering (2017, methodology): “Two types of objective functions are most commonly used in literature — pointwise loss [14, 19] and pairwise loss [27, 33].”
- On the Effectiveness of Linear Models for One-Class Collaborative Filtering (2016, background): “An alternate strategy to adapt matrix factorisation to the implicit feedback setting is the Bayesian Personalised Ranking (BPR) model (Rendle et al. 2009).”
- Session-based Recommendations with Recurrent Neural Networks (2015, neutral): “Although the task can also be interpreted as a classification task, learning-to-rank approaches (Rendle et al., 2009; Shi et al., 2012; Steck, 2015) generally outperform other approaches.”
- Gaussian Ranking by Matrix Factorization (2015, methodology): “Replacing the indicator function by a smooth function immediately leads to a popular training method, Bayesian Personalized Ranking (BPR) [10].”
- Modeling Consumer Preferences and Price Sensitivities from Large-Scale Grocery Shopping Transaction Logs (2017, background): “In practice, we maximize following objective function LLBPR = ∑ i,u,t bi,u(t) ∑ i′ 6=i log pi>i′,u(t) (11) where pi>i′,u(t) = σ(s (prod) i,u (t)− s (prod) i′,u (t)).”
- Ask the GRU: Multi-task Learning for Deep Text Recommendations (2016, background): “We consider the implicit feedback [26, 27] setting, where we only observe whether a person has viewed or liked an item and do not observe explicit ratings. r ij = 1 if user i liked item j and 0 otherwise.”
- A Hybrid Collaborative Filtering Model with Deep Structure for Recommender Systems (2017, methodology): “Similar to (Wang, Wang, and Yeung 2015; Wang and Blei 2011), we use recall as another evaluation metric since the rating information is in the form of implicit feedback (Hu, Koren, and Volinsky 2008; Rendle et al. 2009).”
- A Dynamic Recurrent Model for Next Basket Recommendation (2016, methodology): “In the learning process of DREAM, we adopt Bayesian Personalized Ranking (BPR) [6].”
- Sequential Recommendation with User Memory Networks (2018, background): “•BPR:Bayesian personalized ranking, which is a popularmethod for top-N recommendation [22].”
- Sequential User-based Recurrent Neural Network Recommendations (2017, methodology): “Moreover, using the BPR learning criterion [39], we trained a standard”
- A Neural Influence Diffusion Model for Social Recommendation (2019, methodology): “Instead of directly predicting each user’s point-wise explicit ratings, BPR modeled the pair-wise preferences with the assumption that users prefer the observed implicit feedbacks compared to the unobserved ones [32].”