Abstract: Compared to traditional rating-based collaborative filtering algorithm, the Slope One algorithm is simple and efficient. However, the Slope One algorithm relies on a large number of users' ratings to the item which should be predieted. The rating prediction is affected when users' ratings are not enough and it does not consider the users' habits.For the reason, the semantic similarity of the keywords to describe the items was introduced, which measures the degree of similarity between item-pairs, then a combination of items' semantic similarity and the Slope One algorithm was proposed. Finally,by the standard data set MovieLens, the data of the experiment's result show that the improved algorithm improvers the accuracy of the original algorithm.

Key words: Collaborative filtering, Slope One algorithm, User recommendation, Semantic similarity