With the expand of user and item scales, user-ratings are becoming extremely sparse, and recommender systems use traditional collaborative filtering are facing serious challenges. To ease these problems, many optimized methods combining with items have been put forward, most of which adopt user-ratings to compute item similarity. However, it is noticed that the similarity between items is independent of those subjective ratings, and measuring item similarity through user-ratings is unreasonable. In this paper we propose a new approach to compute item similarity, through mapping each item with a corresponding descriptive document, and computing similarity between document as item similarity, then make basic predictions according to those item similarity to lower sparsity of the user-ratings. After that, classical collaborative filtering steps are taken to generate predictions. Compared with previous methods, the new approach could describe item similarity more objectively and accurately, hence could lower sparsity of user-ratings, experiments also show that the optimized approach could improve accuracy of prediction evidently.
[1]
David Heckerman,et al.
Empirical Analysis of Predictive Algorithms for Collaborative Filtering
,
1998,
UAI.
[2]
Eric Horvitz,et al.
Social Choice Theory and Recommender Systems: Analysis of the Axiomatic Foundations of Collaborative Filtering
,
2000,
AAAI/IAAI.
[3]
John Riedl,et al.
Item-based collaborative filtering recommendation algorithms
,
2001,
WWW '01.
[4]
John Riedl,et al.
An Empirical Analysis of Design Choices in Neighborhood-Based Collaborative Filtering Algorithms
,
2002,
Information Retrieval.
[5]
John Riedl,et al.
Sparsity, scalability, and distribution in recommender systems
,
2001
.