User Modeling and Personalization in the Microblogging Sphere

User Modeling and Personalization in the Microblogging Sphere

Gao, Q.

Houben, G.J.P.M. (promotor)

Electrical Engineering, Mathematics and Computer Science

Software Technology

2013-10-28

Microblogging has become a popular mechanism for people to publish, share, and propagate information on the Web. The massive amount of digital traces that people have left in the microblogging sphere, creates new possibilities and poses challenges for user modeling and personalization. How can microblogging activities be exploited to infer individual users' interests? How can semantically meaningful user profiles be constructed to support different applications? Does the users' microblogging behavior vary between different cultural groups? What is the impact of different user modeling strategies on the characteristics of user profiles and the performance of personalized recommendations? In this thesis, we answer the research questions above and introduce a generic framework that provides a variety of user modeling strategies for inferring individual users' interests from microblogging streams. We propose and evaluate techniques that allow for exploiting external resources to enrich the semantics of short microblogging messages. We explore different approaches for deducing and modeling topics of interests based on enriched microblogging data. Furthermore, we investigate various weighting schemes for constructing user profiles and incorporate temporal constraints into the user modeling process. With flexible design choices, the user modeling framework allows for constructing user profiles which can be consumed in different applications. We apply our user modeling framework to analyze user behavior across cultural groups on microblogging platforms. By exploiting different user characteristics, we unveil key differences in users' microblogging behavior between Chinese and American users. Finally, we analyze and evaluate different user modeling strategies in the context of various personalized recommender systems. The results of our analyses show that the characteristics of user profiles are significantly influenced by different design alternatives. In a set of experiments we reveal that the semantic enrichment of microposts and the consideration of temporal patterns improve the performance of recommender systems. We also prove that the incorporation of public trends and domain specific knowledge into the user modeling process improves the quality of personalized recommendations.

user modeling
personalization
recommender systems
social web
microblog
twitter
sina weibo

https://doi.org/10.4233/uuid:0b691916-b2d6-47a1-a637-d038319d7d7c

9789461862273

Institutional Repository

doctoral thesis

(c) 2013 Gao, Q.