Date of Award

May 2022

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Information Studies

First Advisor

Jin Zhang

Committee Members

Dietmar Wolfram, Iris Xie, Xiangming Mu, Kun Lu

Keywords

health information, information retrieval, information science, recommendation system

Abstract

This study produced a recommendation system that can effectively recommend items on a health portal. Toward this aim, a transaction log that records users’ traversal activities on the Medical College of Wisconsin’s HealthLink, a health portal with a subject directory, was utilized and investigated. This study proposed a mixed-method that included the transaction log analysis method, the Markov chain analysis method, and the inferential analysis method. The transaction log analysis method was applied to extract users’ traversal activities from the log. The Markov chain analysis method was adopted to model users’ traversal activities and then generate recommendation lists for topics, articles, and Q&A items on the health portal. The inferential analysis method was applied to test whether there are any correlations between recommendation lists generated by the proposed recommendation system and recommendation lists ranked by experts. The topics selected for this study are Infections, the Heart, and Cancer. These three topics were the three most viewed topics in the portal. The findings of this study revealed the consistency between the recommendation lists generated from the proposed system and the lists ranked by experts. At the topic level, two topic recommendation lists generated from the proposed system were consistent with the lists ranked by experts, while one topic recommendation list was highly consistent with the list ranked by experts. At the article level, one article recommendation list generated from the proposed system was consistent with the list ranked by experts, while 14 article recommendation lists were highly consistent with the lists ranked by experts. At the Q&A item level, three Q&A item recommendation lists generated from the proposed system were consistent with the lists ranked by experts, while 12 Q&A item recommendation lists were highly consistent with the lists ranked by experts. The findings demonstrated the significance of users’ traversal data extracted from the transaction log. The methodology applied in this study proposed a systematic approach to generating the recommendation systems for other similar portals. The outcomes of this study can facilitate users’ navigation, and provide a new method for building a recommendation system that recommends items at three levels: the topic level, the article level, and the Q&A item level.

Download

Share

COinS