Expertise localization discovered through correlation of key term distribution and community detection in co-author networks

TL;DR

This paper introduces a fast, automated method to identify the research focus of scientific communities in co-authorship networks by combining community detection with key term analysis, optimized for large communities.

Contribution

It presents a novel approach that efficiently determines community topics using bibliographic data and community detection, with a focus on reducing computation time and reliability thresholds.

Findings

Communities with fewer than 8 papers are not reliably characterized.

The method effectively identifies community-specific research topics.

Analysis of community formation factors in co-authorship networks.

Abstract

We present an efficient and effective automatic method for determining the research focus of scientific communities found in co-authorship networks. It utilizes bibliographic data from a database to form the network, followed by fastgreedy community detection to identify communities within large connected components of the network. Text analysis techniques are used to identify community-specific significant terms which represent the topic of the community. In order to greatly reduce computation time, the `Topics' field of each publication in the network is analyzed rather than its entire text. Using this text analysis approach requires a certain level of statistical confidence,therefore analyzing very small communities is not effective with this technique. We find a minimum community size threshold of 8 coauthored papers; below this value, the community's topic cannot be reliably identified with this method. Additionally, we consider the implications this study has regarding factors involved in the formation of scientific communities in co-authorship networks.