Advanced Feature Identification towards Research Article Recommendation: A Machine Learning Based Approach

Abstract

To guide the researcher in searching right paper in the context of an increasing rate of the digital repository of research articles, research paper recommendation system is being advocated highly. To design such a system, crawling, filtering and ranking are the three crucial steps. For the crawling task, the existing approaches consider only direct features, which can be readily obtained from a given paper. Addressing this limitation, this work proposes a novel scheme to define the feature vector representing an article. The proposed work unleashes three hidden features that can not be readily available for a given paper. The three indirect features derived from the direct features itself (which are readily available from a paper) are based on the measurements of keyword diversification, text complexity and citation analysis over time. The rationale behind the proposition of the three indirect features, metrics and their measurements are discussed in detail in this paper. Experimental results clearly substantiate the efficacy of the proposed feature vector.