X. Jin and B. Mobasher (USA)
Collaborative Filtering, Semantic Similarity,Recommendation Systems
Collaborative Filtering (CF) systems address the problem of making personalized recommendation using knowledge discovery techniques. By predicting user ratings on new items based on historical ratings of other users, CF systems can give reasonable recommendations to new users. Traditionally, user-based CF algorithm can give predictions and recommendations by finding similar users. However these algorithms often suffer from scalability, sparsity and first-rater problems. Recently, some researchers have proposed the item-based CF algorithms, which give predictions based on similar items' ratings. These can alleviate the scalability problems, but these algorithms suffer from the sparsity and first-rater problems. In this paper, we present two algorithms which use semantic similarity to enhance item based Collaborative Filtering. In both algorithms, we will extract semantic information about items and compute semantic similarity between them. In the first algorithm, we combine the semantic and rating similarities to find most similar items to a target item. In the second algorithm, we use the combined similarity to fill in the original ratings matrix, and then we run the first algorithm on this less sparse ratings matrix. Our experiments show that both algorithms can achieve better prediction accuracy than traditional item-based CF algorithms. Furthermore, the second algorithm can alleviate the sparsity problem.
Important Links:
Go Back
