Personalized Service-oriented User Profiling And Applications

Personalized services on the Internet can obtain a personalized search results accordingto users’ needs, habits, preference and other characteristics, thus can improve the userexperience and service satisfaction effectively. Nowadays, Internet service providers tend toprovide personalized service, which is regard as the core competitiveness of enterprises. Thebasis of differentiated services is user feature recognition and understanding, and the key ofpersonalized service is to create User Profiles. With the rapid development of CollaborativeTagging Systems, Social Networks and Mobile Internet, users characteristic informationbecome more and more rich. How to create user profiles from different structures and allkinds of information data so as to provide high quality personalized service is a commonresearch goal of academia and industry.In this article, we explore three important problems of personalized service and userprofile construction, including understanding user labeling behavior, identifying users’sentiment tendency of using labels and considering the influence of the social relations inpersonalized service. The contributions of our research work mainly include:(1)In view of the Tag-group effect, we propose User Vector Model based on Tag-group.Then we put forward two methods to create User Profile base on this model. The first one isTag-Group Based User Profiling (TGB) method. The experimental results of personalizedinformation retrieval show that TGB method has better performance on multiple indicesrelative to the baseline algorithms. The second is Tag-Group Integration Based User Profiling(TGIB) method. TGIB filter tag-groups based on the user characteristic expressions. Theexperimental results of recommendation show that TGIB method has better performance onMAE metric than the baseline algorithms.(2)We put forward the label sentiment classification standards and Three-Level VectorModel based on the sentiment polarity of labels. We propose a Three-Level User Profiling(TUP) method to construct user profiles. What’s more, we generalize the Three-Level VectorModel to Multi-Level Vector Model. The experimental results of personalized retrieval showthat TUP method has the best performance in various levels of MUP approach. Besides, TUP method has better performance on multiple indices than other baseline algorithms.(3)We put forward Social Q&A System and its general framework, exploring theproblem form retrieving the relevant information or resources into recommending users whocan provide information or resource. Based on Social Q&A System, we present a RelationshipIntensity Model which is based on the transfer of relationships and a User Model which isbased on social relations. Then we propose a method to construct user profiles in Social Q&ASystem. The experimental results show that this method can discover suitable respondentswho are willing to answer and are familiar with related field, by comparing with thetraditional Q&A System.