User Expertise Modelling Using Social Network Data

Abstract

The ability to understand the expertise of online users is a key component for delivering effective information services such as talent seeking and user recommendation. However, users are often unwilling to make the effort to explicitly provide this information. Data extracted from social networking sites can be a valuable resource for inferring this expertise. But research on user expertise modelling in social networking sites is still immature and faces a number of critical challenges: 1) Modelling the expertise of cold start users; 2) The effective utilisation of a variety of user data for user expertise modelling; 3) The effective application of the modelled user expertise information. The contribution of this thesis is concerned with addressing aspects of these three challenges. Firstly, this research proposes to use the static social media profile of a cold start user to model their language information. This is mainly based on the intuition that a user's experiences could imply what languages they know. A language and social relation-based factor graph model is proposed which exploits the dependency relations between languages as well as social relations between profiles to better model this problem. Secondly, this research focuses on the popular micro-blogging site Twitter and aims to model the general topics of expertise a user has knowledge of, based upon their various user data. Specifically, sentiment analysis is first used to assess the importance of each tweet, the primary data of a Twitter user, in user expertise modelling. Then, based on discriminative learning, this research proposes a model that infers a user's expertise under the collective but discriminative influence of various data related to the user. The proposed model exploits the dependency relations between expertise topics and consistency relations between different types of user data in order to better infer the user?s topical expertise. Thirdly, this research studies the application of the proposed user expertise modelling methods in the scenario of a community question answering site. It aims to model the expertise information of more users in the platform through the adoption of the proposed methods and help to more effectively find potential answerers to newly posted questions, which offers an effective way to improve the question answering service.