A spatio-temporal approach to collaborative filtering

Abstract

In this paper, we propose a novel spatio-temporal model for collaborative filtering applications. Our model is based on low-rank matrix factorization that uses a spatio-temporal filtering approach to estimate user and item factors. The spatial component regularizes the factors by exploiting correlation across users and/or items, modeled as a function of some implicit feedback (e.g., who rated what) and/or some side information (e.g., user demographics, browsing history). In particular, we incorporate correlation in factors through a Markov random field prior in a probabilistic framework, whereby the neighborhood weights are functions of user and item covariates. The temporal component ensures that the user/item factors adapt to process changes that occur through time and is implemented in a state space framework with fast estimation through Kalman filtering. Our spatio-temporal filtering (ST-KF hereafter) approach provides a single joint model to simultaneously incorporate both spatial and temporal structure in ratings and therefore provides an accurate method to predict future ratings. To ensure scalability of ST-KF, we employ a mean-field approximation for inference. Incorporating user/item covariates in estimating neighborhood weights also helps in dealing with both cold-start and warm-start problems seamlessly in a single unified modeling framework; covariates predict factors for new users and items through the neighborhood. We illustrate our method on simulated data, benchmark data and data obtained from a relatively new recommender system application arising in the context of Yahoo! Front Page.

Enhanced Cross Domain Recommender System using Contextual parameters in Temporal Domain

“…[4], [5] and [6] deal with different methods of cross domain collaborative filtering with temporal domain. In [4], the temporal domains are considered.…”

“…In both factorization and neighborhood models, the inclusion of temporal dynamics proved very useful in improving quality of predictions, more than various algorithmic enhancements. [6] uses derived method based on a Bayesian latent factor model which can be inferred using Gibbs sampling. This deals with user interest drift over time in single domain.…”

“…Collaborative filtering with temporal dynamics [5] Yes No A spatio-temporal approach to collaborative filtering [6] Yes No…”

Daily-Aware Personalized Recommendation based on Feature-Level Time Series Analysis

“…Oku [20] and Yuan [37] integrated time factors into the similarity calculation of neighbour-hood based methods for time-aware CF; Koren [16] proposed timeSVD++ for dynamic CF with factor models by tracking the drifting user/item biases across different time bins; Xiang et al [34] further extended the dynamic modeling to implicit feedback through random walk on sessions-based temporal graphs; Lu [19] and Shi [25] adapted the famous Matrix Factorization (MF) [28] approaches for time-aware CF, while Karatzoglou [15] and Gantner [13] leveraged tensor factorization to integrate the many contextual features. The topic modeling approaches are also investigated by Chen [7] and Vaca [30], while recently, Wang et al [32,31] borrowed the opportunity models from survival analysis to predict users' subsequent purchases for time-aware recommendation.…”

“…Koren [16] and Xiang [34] adopted the approaches of dynamic modeling through time bins on explicit feedbacks, and sessions on implicit feedbacks, respectively, which were further combined by Dror [11] into Yahoo! Music recommendation and achieved superior results in KddCup-2011; At the same time, Lu [19] and Shi [25], as well as Karatzoglou [15] and Gantner [13] adopted matrix/tensor factorization to model time as context information. Recently, Wang [32,31] investigated users' subsequent purchasing behavior with opportunity models for recommendation.…”