Proposal on matrix-based collaborative filtering using personal values
Yuya Shiraishi, Yasufumi Takama · 2017
This paper proposes a matrix-based collaborative filtering based on users' personal values. On many online shopping sites, hotel booking sites, etc, a huge amount of items are available, from witch it becomes difficult for users to find items of interest. Recommender systems help users by suggesting items expected to be preferred by them. One of the common and successful recommendation methods is collaborative filtering, which determines items to recommend based on information about users' past behaviors. Among various kinds of collaborative filtering proposed so far, matrix factorization-based approach can make a good recommendation especially for sparse rating matrix. On the other hand, it is generally known that personal values affect user's decision making on evaluating / purchasing items, and recommender systems which try to make use of personal values have been studied. This paper proposes a matrix-based collaborative filtering that uses personal values. In the proposed method, user models and item models reflecting personal values are represented as matrices, and predicted score of a target item is calculated by taking the product of them. Effectiveness of the proposed method is evaluated with movie dataset. Experimental results show that the proposed method achieves comparable results with state-of-the-art matrix-based recommendation methods. In particular, it is observed that the proposed method outperforms the existing methods for users with less rating histories. It is also shown that the proposed recommender system is effective for recommending long-tail items.