Cluster validation techniques for Bibliographic databases

Abstract

In entity name disambiguation technique, records of same entity are clustered together. One of the major challenges in such technique is to validate the result as the actual or correct results are often not known or difficult to know. In this context, three commonly known evaluation measures are precision, recall and f-measure. All these indices are external validity indices as they all need gold standard data. But in Bibliographic databases like DBLP, Arnetminer, Scopus, Web of Science etc., obtaining golden standard is very difficult for each entity. So, there is a need to use some other metrics to evaluate the performance on Bibliographic data. In this paper, a novel scheme based on internal validity index is used to evaluate the performance of entity name disambiguation algorithm. Several distance measures are used here to compute the similarity between two records. These functions are then incorporated in the definitions of internal validity indices.