The full text of this item is not available at this time because the student has placed this item under an embargo for a period of time. The Libraries are not authorized to provide a copy of this work during the embargo period, even for Texas A&M users with NetID.
Personalized Expert Recommendation: Models and Algorithms
MetadataShow full item record
Many large-scale information sharing systems including social media systems, questionanswering sites and rating and reviewing applications have been growing rapidly, allowing millions of human participants to generate and consume information on an unprecedented scale. To manage the sheer growth of information generation, there comes the need to enable personalization of information resources for users — to surface high-quality content and feeds, to provide personally relevant suggestions, and so on. A fundamental task in creating and supporting user-centered personalization systems is to build rich user profile to aid recommendation for better user experience. Therefore, in this dissertation research, we propose models and algorithms to facilitate the creation of new crowd-powered personalized information sharing systems. Specifically, we first give a principled framework to enable personalization of resources so that information seekers can be matched with customized knowledgeable users based on their previous historical actions and contextual information; We then focus on creating rich user models that allows accurate and comprehensive modeling of user profiles for long tail users, including discovering user’s known-for profile, user’s opinion bias and user’s geo-topic profile. In particular, this dissertation research makes two unique contributions: First, we introduce the problem of personalized expert recommendation and propose the first principled framework for addressing this problem. To overcome the sparsity issue, we investigate the use of user’s contextual information that can be exploited to build robust models of personal expertise, study how spatial preference for personally-valuable expertise varies across regions, across topics and based on different underlying social communities, and integrate these different forms of preferences into a matrix factorization-based personalized expert recommender. Second, to support the personalized recommendation on experts, we focus on modeling and inferring user profiles in online information sharing systems. In order to tap the knowledge of most majority of users, we provide frameworks and algorithms to accurately and comprehensively create user models by discovering user’s known-for profile, user’s opinion bias and user’s geo-topic profile, with each described shortly as follows: —We develop a probabilistic model called Bayesian Contextual Poisson Factorization to discover what users are known for by others. Our model considers as input a small fraction of users whose known-for profiles are already known and the vast majority of users for whom we have little (or no) information, learns the implicit relationships between user?s known-for profiles and their contextual signals, and finally predict known-for profiles for those majority of users. —We explore user’s topic-sensitive opinion bias, propose a lightweight semi-supervised system called “BiasWatch” to semi-automatically infer the opinion bias of long-tail users, and demonstrate how user’s opinion bias can be exploited to recommend other users with similar opinion in social networks. — We study how a user’s topical profile varies geo-spatially and how we can model a user’s geo-spatial known-for profile as the last step in our dissertation for creation of rich user profile. We propose a multi-layered Bayesian hierarchical user factorization to overcome user heterogeneity and an enhanced model to alleviate the sparsity issue by integrating user contexts into the two-layered hierarchical user model for better representation of user’s geo-topic preference by others.
Lu, Haokai (2017). Personalized Expert Recommendation: Models and Algorithms. Doctoral dissertation, Texas A & M University. Available electronically from