Completing features for author name disambiguation (AND): an empirical analysis

This study presents a feature enriched AND dataset to develop diverse and better performance achieving AND techniques, by utilizing AND features which have better discriminating abilities to solve this problem. Current AND datasets have limited number of useful AND features in them, some of them have been curated keeping in mind specific scenarios or contexts and some of them are domain specific. Rather than limiting the labelled datasets to be domain specific, contextual or hold limited feature values, it is better to leave their usage limit as a choice with respect to the technique which is trying to solve this problem. In this paper, our proposed labelled dataset “CustAND” provides a set of 7886 publication records, where each record covers more than eleven useful features values. The dataset covers multi domains as well as different ethnical group authors. CustAND is collected from multiple web sources, where raw data is extracted from digital libraries and search engines. This data is later cross checked, hand labelled and confirmed (authorship confirmation) by a team of graduate students with 100% accuracy. The raw data after pre-processing is validated by checking author’s personal web pages, different profile pages, their affiliations, and emails. This new dataset complements the availability of useful feature values which are crucial in developing generic and better performance achieving techniques to solve the author’s name ambiguity problem generally faced by the digital libraries.