A novel recommendation model with Google similarity

摘要

Previous studies on collaborative filtering have mainly adopted local resources as the basis for conducting analyses, and user rating matrices have been used to perform similarity analysis and prediction. Therefore, the efficiency and correctness of item-based collaborative filtering completely depend on the quantity and comprehensiveness of data collected in a rating matrix. However, data insufficiency leads to the sparsity problem. Additionally, cold-start is an inevitable problem concerning with how local resources are used as the basis for conducting analyses. This paper proposes a new idea by identifying an additional database to support item-based collaborative filtering. Regardless of whether a recommender system operates under a normal condition or applies a sparse matrix and introduces new items, this extra database can be used to accurately calculate item similarity. Moreover, prediction results acquired from two distinctive sets of data can be integrated to enhance the accuracy of the final prediction or successful recommendation.