Improving Latent Factor Models via Personalized Feature Projection for One Class Recommendation
Authors:
Open Access
- 17 October 2015
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 821-830
Abstract
Latent Factor models, which transform both users and items into the same latent feature space, are one of the most successful and ubiquitous models in recommender systems. Most existing models in this paradigm define both users' and items' latent factors to be of the same size and use an inner product to represent a user's "compatibility" with an item. Intuitively, users' factors encode "preferences" while item factors encode "properties", so that the inner product encodes how well an item matches a user's preferences. However, a user's opinion of an item may be more complex, for example each dimension of each user's opinion may depend on a combination of multiple item factors simultaneously. Thus it may be better to view each dimension of a user's preference as a personalized projection of an item's properties so that the preference model can capture complex relationships between items' properties and users' preferences. Therefore, in this paper we propose a novel personalized feature projection method to model users' preferences over items. Specifically, for each user, we define a personalized projection matrix, which takes the place of user-specific factors from existing models. This matrix describes a mapping between items' factors and users' preferences in order to build personalized preference models for each user and item. The proposed personalized feature projection method is quite general and existing latent factor models, for example, can be cast as a special case. We present three objective functions to optimize predictions in the form of ranked lists of users' preferences over items, and demonstrate how each can be used to improve one-class recommendation performance. Experiments are conducted on four real-world datasets and our results show that our personalized feature projection method outperforms several state-of-the-art methods on various evaluation metrics.Keywords
Funding Information
- National Grand Fundamental Research 973 Program of China (No. 2014CB340405)
- Microsoft Research Asia Regional Seed Fund in Big Data Research (Grant No. FY13-RES-SPONSOR-036)
- Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. CUHK 413213)
This publication has 32 references indexed in Scilit:
- Learning to rank recommendations with the k-order statistic loss2013
- MyMediaLite2011
- Collaborative competitive filtering2011
- User Graph Regularized Pairwise Matrix Factorization for Item Recommendation2011
- Advances in Collaborative Filtering2010
- Large scale image annotation: learning to rank with joint word-image embeddingsMachine Learning, 2010
- Improving maximum margin matrix factorizationMachine Learning, 2008
- Information-theoretic metric learning2007
- TECHNICAL NOTE. AN EMPIRICAL MODIFICATION TO LINEAR WAVE THEORY.Proceedings of the Institution of Civil Engineers, 1976
- On Information and SufficiencyThe Annals of Mathematical Statistics, 1951