Repository logo
 

IMPROVING COLLABORATIVE FILTERING RECOMMENDER BY USING MULTI-CRITERIA RATING AND IMPLICIT SOCIAL NETWORKS TO RECOMMEND RESEARCH PAPERS

Date

2016-09-21

Journal Title

Journal ISSN

Volume Title

Publisher

ORCID

Type

Thesis

Degree Level

Doctoral

Abstract

Research paper recommender systems (RSs) aim to alleviate the information overload of researchers by suggesting relevant and useful papers. The collaborative filtering in the area of recommending research papers can benefit by using richer user feedback data through multi-criteria rating, and by integrating richer social network data into the recommender algorithm. Existing approaches using collaborative filtering or hybrid approaches typically allow only one rating criterion (overall liking) for users to evaluate papers. We conducted a qualitative study using focus group to explore the most important criteria for rating research papers that can be used to control the paper recommendation by enabling users to set the weight for each criterion. We investigated also the effect of using different rating criteria on the user interface design and how the user can control the weight of the criteria. We followed that by a quantitative study using a questionnaire to validate our findings from the focus group and to find if the chosen criteria are domain independent. Combining social network information with collaborative filtering recommendation algorithms has successfully reduced some of the drawbacks of collaborative filtering and increased the accuracy of recommendations. All existing recommendation approaches that combine social network information with collaborative filtering in this domain have used explicit social relations that are initiated by users (e.g. “friendship”, “following”). The results have shown that the recommendations produced using explicit social relations cannot compete with traditional collaborative filtering and suffer from the low user coverage. We argue that the available data in social bookmarking Web sites can be exploited to connect similar users using implicit social connections based on their bookmarking behavior. We explore the implicit social relations between users in social bookmarking Web sites (such as CiteULike and Mendeley), and propose three different implicit social networks to recommend relevant papers to users: readership, co-readership and tag-based implicit social networks. First, for each network, we tested the interest similarities of users who are connected using the proposed implicit social networks and compare them with the interest similarities using two explicit social networks: co-authorship and friendship. We found that the readership implicit social network connects users with more similarities than users who are connected using co-authorship and friendship explicit social networks. Then, we compare the recommendation using three different recommendation approaches and implicit social network alone with the recommendation using implicit and explicit social network. We found that fusing recommendation from implicit and explicit social networks can increase the prediction accuracy, and user coverage. The trade-off between the prediction accuracy and diversity was also studied with different social distances between users. The results showed that the diversity of the recommended list increases with the increase of social distance. To summarize, the main contributions of this dissertation to the area of research paper recommendation are two-fold. It is the first to explore the use of multi-criteria rating for research papers. Secondly, it proposes and evaluates a novel approach to improve collaborative filtering in both prediction accuracy (performance) and user coverage and diversity (nonperformance measures) in social bookmarking systems for sharing research papers, by defining and exploiting several implicit social networks from usage data that is widely available.

Description

Keywords

recommender system, collaborative filtering, social recommendation, user coverage, diversity

Citation

Degree

Doctor of Philosophy (Ph.D.)

Department

Computer Science

Program

Computer Science

Citation

Part Of

item.page.relation.ispartofseries

DOI

item.page.identifier.pmid

item.page.identifier.pmcid