Article Recommendation In Research Social Networks

The rapid development of information technology has increased the amount of available information, which currently poses significant challenges for users who seek relevant information. As emerging tools to deal with information overload, recommendation techniques are critically important in providing personalized online information services. With the prevalence of research social networks, determining effective methods for recommending scientific articles to online scholars has become a challenging and complex task.Current studies on article recommendation works are focused on digital libraries and reference sharing websites while studies on research social networking websites have seldom been conducted. Existing content-based approaches place emphasis on direct keyword weighting methods which often cause keyword mismatch problems. Behavior connections have been widely used for collaborative filtering approaches, but additional information has not been deeply mined. The quality information of articles has been largely ignored in previous studies, thus raising the need for a unified recommendation framework that combines relevance, connectivity and quality features.With the goal of addressing the identified challenges and research gaps, this thesis presents an integrated recommendation framework that combines relevance, connectivity and quality analyses for article recommendation in academic contexts. The relevance analysis module employs a semantic content filtering method to address the keyword mismatch problem. The connectivity analysis module utilizes scoring-based and random walk with start based methods to alleviate the data sparsity problem. The quality analysis module is proposed and incorporated into the integrated recommendation model to improve performance. The effectiveness of the proposed framework and methods is verified using an offline experiment and a user study on CiteULike and Scholarmate, respectively. The results demonstrate that our proposed methods outperform existing baselines.The proposed algorithm and designed recommender system are incorporated into existing research social networking websites to facilitate content sharing and potential collaboration. Thus, the algorithm and system can be generalized to other personalization applications in other social media websites.