User Interest Expansion Based On Community Detection

With the rapid development of hardware, software and Internet technology, toget the most valuable information in the shortest time for most person who are in thefast pace of life is undoubtedly the most important problem to be solved. In recentyears, pattern of the user behavior on the network quietly changed from the previousfinding the information you need initiatively to passively accept the recommendationof the news now, therefore, research on the "personalized recommendations" is still ahot area today. Traditional personalized recommendation algorithm only took entitywhich has been searched for a specific person (such as a star), movies, novels,merchandise and other information into account in the interest modeling stage, thereis no prediction on the change of the user interest. When there is no relevant newsabout the interest to the user, or less related link to high-quality web pages, we canrecommend to the user the results of extension.We are inspired by the idea of community detection, assembling the close-knitindividuals into a community, separating the rare-connection individuals formdifferent communities. To this article on the user interest expansion, we assemble theremarkably similar user interest directions into a group, separating therare-connection user interest directions form different communities. The remarkablysimilar user interest directions are in the same interest groups, the results ofexpansion come from this group.This paper presents a model of user interest expansion based on communitydiscovery, aimed at discovering the user groups with similar interests, expandingusers’ original interests direction by different interests direction of others in thegroup. Model is divided into three modules: firstly, check user search logs and clicklogs, find points of interest that users are most interested in, complete the modeling of users’ interests; and then calculate the degree of similarity between the directionsof users’ interests. We consider the association relationship from the direction ofsemantic similarity between user interest literally, on the relevance language model,implicit association between query and title, so that the relationship betweendirection of the user’s interest is more comprehensive and scientific; and finally,using the method of community detection, according to relationship of the interestentity, and the direction of interest, identify groups with similar interests, extend theuser interest direction based on user interests of others within the groupappropriately.As can be seen in the experiment, user interest expansion is not to beunderestimated for personalized recommendation, with nearly doubled growth.Compared with the common K-means clustering algorithm, the same rule applied toextend, the algorithm in this paper is more effective.