Abstract:

Collaborative filtering is a widely used recommendation algorithm, but problems such as low efficiency and data sparseness still exist. In order to solve these problems, we present an improved clustering recommendation algorithm. The algorithm introduces a cloud model, in which the expectation, entropy and hyper entropy are calculated according to the item attributes and user attributes dimensions. To build up a user interest model, the influence of rating time, rating level and rating habits are also taken into account. Then the similarities of user interests are compared by the corrected similarity measurement based on cloud model, and the Kmeans algorithm is adopted to perform clustering. Finally, the recommendation results of the public projects are merged by using the proportion of the participants who will make predictions. Experiment results on the MovieLens show that the algorithm can not only solve the problem of low efficiency and data sparseness but also improve the accuracy of the recommendation results.

Key words: collaborative filtering;cloud model;clustering;data sparseness