Date

2014

Document Type

Dissertation

Degree

Doctor of Philosophy

Department

Computer Science

First Adviser

Davison, Brian D.

Other advisers/committee members

Chuah, Mooi Choo; Heflin, Jeff; Nagel, Roger; Ridings, Catherine

Abstract

Social Network research has attracted the interests of many researchers, not only in analyzing the online social networking applications, such as Facebook and Twitter, but also in providing comprehensive services in scientific research domain. We define an Academic Network as a social network which integrates scientific factors, such as authors, papers, affiliations, publishing venues, and their relationships, such as co-authorship among authors and citations among papers. By mining and analyzing the academic network, we can provide users comprehensive services as searching for research experts, published papers, conferences, as well as detecting research communities or the evolutions hot research topics. We can also provide recommendations to users on with whom to collaborate, whom to cite and where to submit.In this dissertation, we investigate two main tasks that have fundamental applications in the academic network research. In the first, we address the problem of expertise retrieval, also known as expert finding or ranking, in which we identify and return a ranked list of researchers, based upon their estimated expertise or reputation, to user-specified queries. In the second, we address the problem of research action recommendation (prediction), specifically, the tasks of publishing venue recommendation, citation recommendation and coauthor recommendation. For both tasks, to effectively mine and integrate heterogeneous information and therefore develop well-functioning ranking or recommender systems is our principal goal. For the task of expertise retrieval, we first proposed or applied three modified versions of PageRank-like algorithms into citation network analysis; we then proposed an enhanced author-topic model by simultaneously modeling citation and publishing venue information; we finally incorporated the pair-wise learning-to-rank algorithm into traditional topic modeling process, and further improved the model by integrating groups of author-specific features. For the task of research action recommendation, we first proposed an improved neighborhood-based collaborative filtering approach for publishing venue recommendation; we then applied our proposed enhanced author-topic model and demonstrated its effectiveness in both cited author prediction and publishing venue prediction; finally we proposed an extended latent factor model that can jointly model several relations in an academic environment in a unified way and verified its performance in four recommendation tasks: the recommendation on author-co-authorship, author-paper citation, paper-paper citation and paper-venue submission. Extensive experiments conducted on large-scale real-world data sets demonstrated the superiority of our proposed models over other existing state-of-the-art methods.

Share

COinS