Abstract:

The recommendation methods based on network structure have the problem of over-recommending “popular resources” and ignoring “unpopular resources”. However, in practice, “unpopular resources” are more in line with users’ personalized requirements. Therefore, improving the novelty of the recommendation results by mining “unpopular resources” has become one of the research directions of the recommendation algorithms. This paper improves the recommendation method based on the network structure to improve the recommendation of “unpopular resources”. Firstly, a user-projects-tags network structure is constructed. Secondly, the users’ rating difference of the item is used as the weight of the user-projects energy transfer so as to improve the recommendation accuracy, and the information entropy of the tags is used as the weight of the projects-tags energy transfer to increase the recommendation of “unpopular resource”. Finally, the recommendation list is generated by linear weighting. Experimental results on the Movielens dataset prove that the proposed algorithm takes into account the accuracy of the recommendation results and improves the recommendation of “unpopular resources” in the recommendation results.

Key words: network structure, unpopular resource, information entropy, personalized recommendation