Accuracy of inter-researcher similarity measures based on topical and social clues

  • Authors:
  • Guillaume Cabanac

  • Affiliations:
  • Computer Science Department, IRIT UMR 5505 CNRS, University of Toulouse, Toulouse Cedex 9, France 31062

  • Venue:
  • Scientometrics
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Scientific literature recommender systems (SLRSs) provide papers to researchers according to their scientific interests. Systems rely on inter-researcher similarity measures that are usually computed according to publication contents (i.e., by extracting paper topics and citations). We highlight two major issues related to this design. The required full-text access and processing are expensive and hardly feasible. Moreover, clues about meetings, encounters, and informal exchanges between researchers (which are related to a social dimension) were not exploited to date. In order to tackle these issues, we propose an original SLRS based on a threefold contribution. First, we argue the case for defining inter-researcher similarity measures building on publicly available metadata. Second, we define topical and social measures that we combine together to issue socio-topical recommendations. Third, we conduct an evaluation with 71 volunteer researchers to check researchers' perception against socio-topical similarities. Experimental results show a significant 11.21% accuracy improvement of socio-topical recommendations compared to baseline topical recommendations.