Research Of Similarity Weight Computation In Neighborhood-based Collaborative Filtering Recommendation Systems

Recommender systems have proven to be valuablemeans for online users to cope with the information overload and have become one of the most powerful and popular tools in electronic commerce. Collaborative filtering is considered to be the most popular and widely implemented technique in recommender systems.Neighborhood-based methods for collaborative filtering enjoy considerable popularity due to their simplicity, efficiency,and their ability to produce accurate and personalized recommendations. This approach recommends to the active user the items that other users with similar tastes liked in the past. The similarity in taste of two users is calculated based on the similarity in the rating history of the users. Because rating correlation measures the similarity between two users by comparing their ratings for the same items, users can be neighbors only if they have rated common items.The similarity weights play a double role in neighborhood-based recommendation methods: 1) they allow the selection of trusted neighbors whose ratings are used in the prediction, and 2) they provide the means to give more or less importance to these neighbors in the prediction. The computation of the similarity weights is one of the most critical aspects of building a neighborhood-based recommender system,as it can have a significant impact on both its accuracy and its performance.This paper focuses on the research of how to get the available similarity between two users or items and how to compute similarity weights accurately, the main contents and contributions include:(1)The computation of the similarity weights is one of the most critical aspects of building a neighborhood-based recommender system,as it can have a significant impact on both its accuracy and its performance. Traditional similarity computation methods suffer from rating data sparsity problem, and can only obtain the local similarity by measuring the common ratings. Thus, this paper propose a novel similarity computation method—JS, which can be employed to compute similarities by combined local similarity and global similarity. And an improved method is proposed to optimize the relation between local similarity and global similarity. The two methods can keep the simplicity and efficiency without additional complexity.Experimental results show that the new methods outperform traditional methods or common significance weighting methods on the prediction of ratings.(2)When the rating data is sparse, two users or items are likely to have no common ratings, and consequently neighborhood-based approaches will predict ratings using none of neighbors. A solution for the problem is to try to find the third user or item who is both similar to the two users or items based on the proverb "The friend of my friend is my friend". A novel transitive similarity is proposed to compute the similarity weights of the two users or items indirectly. The approach is not only used to measure the similarity without common ratings, but also used to revise the inaccurate similarity weights with few ratings. Experiments show that this method can provide more recommendations.(3) Traditional similarity computation methods usually are correlation-based similarity. Correlations have positive correlation and negative correlation. In general,negative correlations are less reliable than positive ones. Intuitively, this is because strong positive correlation between trousers is a good indicator of their belonging to a common group. Although negative correlation may indicate membership to different groups, it does not tell how different these groups are, or whether these groups are compatible for other categories of items. In most cases similarities always can be discarded. Based on the proverb "The enemy of my enemy is my friend", negative correlations can be used to find the enemy and the enemy of the one. A novel similarity computation method based on negative correlations proposed to compute the similarity weights of the two users or items indirectly. The experiment results demonstrate that the approach can provide more recommendations...