An Evaluation of Structure Based Similarity Indexes for Collaborative Filtering

Recommender systems are significant for web based e-commerce, and collaborative filtering (CF) algorithms are proved to be the most successful recommendation techniques. For CF algorithms, the similarity index is the heart, and many indexes have been proposed. Recently,with the development of complex networks, the structure based similarity indexes have become a hot topic . In this paper, we implemented six structure based local similarity indexes, i.e., salton index, jaccard index, adamic-adar index, common neighbors, sorensen index, and RA index, compared with two common used benchmark indexes, i.e., cosine index and Pearson correlation coefficient , using seven evaluation metrics, i.e., mean absolute error, rooted mean square error, Precise, Recall, F1, diversity and novelty and half-life utility. Experimental results demonstrate that the structural based similarity indexes outperform the benchmark indexes.