Temporal Collaborative Filtering with Bayesian Probabilistic Tensor Factorization

By Liang Xiong, X. Chen, Tzu-Kuo Huang, Jeff Schneider, Jaime G. Carbonell. 2010

731 citations recorded by the source corpus.

Real-world relational data are seldom stationary, yet traditional collaborative filtering algorithms generally rely on this assumption. Motivated by our sales prediction problem, we propose a factor-based algorithm that is able to take time into account. By introducing additional factors for time, we formalize this problem as a tensor factorization with a special constraint on the time dimension. Further, we provide a fully Bayesian treatment to avoid tuning parameters and achieve automatic model complexity control. To learn the model we develop an efficient sampling procedure that is capable of analyzing large-scale data sets. This new algorithm, called Bayesian Probabilistic Tensor Factorization (BPTF), is evaluated on several real-world problems including sales prediction and movie recommendation. Empirical results demonstrate the superiority of our temporal model.

Subfields: Artificial Intelligence, Computational Mathematics, Information 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

In Computational Mathematics

In Information Systems

Citations in the corpus

From amateurs to connoisseurs: modeling the evolution of user expertise through online reviews (2013, background): “Better performance may be obtained by modeling large-scale global changes at the level of entire communities [41], or by developing separate models for short term changes (e.g. due to external events), and long-term trends [40].”
Time-aware recommender systems: a comprehensive survey and analysis of existing evaluation protocols (2013, background): “Another example of a continuous time-based model is given in Xiong et al. (2010), where a Bayesian probabilistic tensor factorization (TF) model is proposed.”
Recommending Groups to Users Using User-Group Engagement and Time-Dependent Matrix Factorization (2016, methodology): “Other authors have examined time-dependent methods such as tensor factorization, session-based temporal graph model and dynamic matrix factorization etc. (Xiong et al. 2010; Xiang et al. 2010; Koenigstein, Dror, and Koren 2011; Lathia, Hailes, and Capra 2009; Chua, Oentaryo, and Lim 2013).”
Tensors for Data Mining and Data Fusion (2016, background): “One of the ﬁrst attempts to apply tensors to collaborative ﬁltering and recommendation systems is Xiong et al. [2010].”
Tensor Decomposition for Signal Processing and Machine Learning (2016, background): “in [96] from a Bayesian point of view, which also proposed a probabilistic model for the hyper-parameters coupled with Markov Chain Monte-Carlo (MCMC) techniques for automated parameter tuning.”
Fast context-aware recommendations with factorization machines (2011, methodology): “In the second main part, we propose a new fast alternating least square (ALS) optimization algorithm that makes FM much easier applicable compared to SGD algorithms because it works without any learning rate.”
TFMAP: optimizing MAP for top-n context-aware recommendation (2012, methodology): “Two state-of-theart approaches have been proposed to date, one based on tensor factorization [14, 35] and the other on factorization machines [25] (FM).”
Neural Survival Recommender (2017, background): “Among those works, temporal dynamics was one of the main themes since after all, users’ behavior change over time and it is quite crucial that the model is able to capture such eﬀects [22, 33].”
Relational Representation Learning for Dynamic (Knowledge) Graphs: A Survey (2019, background): “Xiong et al. [236] propose a probabilistic factorization of A where the nodes are represented as normal distributions with the means coming from a k s and b k s.”
Hinge-Loss Markov Random Fields and Probabilistic Soft Logic (2015, methodology): “Based on settings used by Xiong et al. (2010), we set the rank of the decomposition to 30 and use 100 iterations of burn in and 100 iterations of sampling.”
Collaborative Filtering beyond the User-Item Matrix (2014, methodology): “Probabilistic tensor factorization (PTF), an approach that combines PMF and TF, was developed and employed with the CP model for incorporating time information for time-dependent CF [Xiong et al. 2010], and with the Tucker model for the purpose of review recommendation [Moghaddam et al. 2012].”
Discovering Attribute Shades of Meaning with the Crowd (2015, methodology): “See the Bayesian Probabilistic Tensor Factorization (BPTF) approach of (Xiong et al, 2010) for more details.”
Streaming Recommender Systems (2016, methodology): “In recent years, there is a large literature exploiting temporal information [8, 13, 25, 26] and it is evident that explicitly modeling temporal dynamics greatly improves the recommendation performance [13, 24].”
Word Semantic Representations using Bayesian Probabilistic Tensor Factorization (2014, methodology): “Our model is a variation of the BPMF model (Salakhutdinov and Mnih, 2008), and is similar to the temporal BPTF model (Xiong et al., 2010).”
Time-Sensitive Recommendation From Recurrent User Activities (2015, methodology): “We compare with the stateof-art method [3] which considers poisson regression as the loss function to fit the number of events in each discretized time slot and shows better performance compared to other alternatives with the squared loss [25, 13, 22, 21].”
Predicting the Next Location: A Recurrent Model with Spatial and Temporal Contexts (2016, methodology): “Tensor Factorization (TF) (Xiong et al. 2010) treats time bins as another dimension and generate latent vectors of users, items and time bins via factorization.”