Emerging Scientific Topic Discovery by Analyzing Reliable Patterns of Infrequent Synonymous Biterms

Abstract

Emerging scientific topics are those topics that the number of related articles was small in the past but has grown dramatically in recent years. Automatic discovery of emerging scientific topics has become increasingly necessary because of the exponentially increasing of research papers. Such discovery enables broad applications, such as optimizing resource allocations for promising research areas, predicting future technology trends, finding knowledge gaps and new concepts, and recommending personalized research directions. In this paper, we provide a framework of emerging topic discovery methods using Infrequent Synonymous Biterm (ISB), which automatically extracts the dedicated knowledge from the infrequent patterns of synonymous biterms in a corpus (e.g., paper titles); each term in a synonymous biterm represents a collaborating supertopic, whose collaboration originates an emerging topic. In particular, we propose an Analyzing Reliable Patterns of Infrequent Synonymous Biterms (ARPISB) method, which guarantees the quality of the result emerging topics by adaptively giving larger weights to more reliable ISB. Extensive experiments on five subfields' scholarly papers demonstrate the significant and robust improvement of the accuracy of emerging scientific topic discovery.