Research On The Methods Of Mining User Interests In Social Networks

With the development of social networks,researches on mining user interests in the fields of marketing,social science,and cyberspace governance are becoming more and more important.In order to study the intrinsic factors and extrinsic manifestations of user interests more comprehensively and precisely,in this dissertation we analyses user interest characteristics from the perspective of individual and group users in combination with the features of social networks,focusing on mining user interests from the aspects of text content,social relations,static community and dynamic interaction.The main contributions are summarized as follows.1.A text-based algorithm for interests mining is proposed to tackle the issue of short texts and scattering topics.This algorithm extracts interest keywords,and mines individual interests by mapping these keywords onto an interest hierarchy built on external knowledge base and news websites.We also propose an improved semisupervised algorithm,which is able to identify and filter topic noise in large amounts of data with a small amount of manually annotated data.2.A relation-based algorithm for capturing user interests in a large-scale heterogeneous network is proposed.This algorithm analyses the features of current social networks,and uses different approaches to model super nodes such as celebrities and for average users.By introducing the label propagation algorithm to calculate the spread of interest topics among average users,we rapidly build a relatively comprehensive user interest map in a huge heterogeneous network.3.An algorithm identifying user communities is proposed,which combines user similarities and static relations to find communities.Then the interest characteristic of community is analyzed by supervised learning.We focus on the selection of user tags,network structure and other attributes to calculate user similarity before mapping the similarity to user relations,and use edge clustering algorithm to find multiple communities.Finally,the common interest of the community is learned from some annotated users of this community.4.An algorithm for mining the interests of interaction groups is proposed.It builds a strategic model of user behaviour with game theory to identify dynamic interaction groups and mines group interests by mapping interactive topics to a interest hierarchy.It models user motivation in the game of strategy,by defining the utility function of individual users based on interests,and introducing local equilibrium instead of global equilibrium.Since interaction groups are highly topic-related,the interest of the group is easily obtained while the group is identified.Finally,a prototype system for analyzing social network users is developed,which provides functions for modeling and describing individual user interest characteristics,as well as analyzing and tracking interests and behaviors of specific groups.Some modules of the system have been deployed in several national projects.The system features the analysis and prediction of behavioural characteristics such as interests of specific users and large-scale user groups in a real network environment,providing technical support for the state administration on areas such as online analysis of public opinion,hot news analysis and forewarning,etc.