Investigating the characteristics and research impact of sentiments in tweets with links to computer science research papers

Abstract

Research papers are often shared in Twitter to facilitate better readership. Tweet counts are embedded in journal websites and academic databases, to emphasize the impact of papers in social media. However, more number of tweets per paper is doubted as an indicator of research quality. Hence, there is a need to look at the intrinsic factors in tweets. Sentiment is one of such factors. Earlier studies have shown that neutral sentiment is predominantly found in tweets with links to research papers. In this study, the main intention was to have a closer look at the non-neutral sentiments in tweets to understand whether there is some scope for using such tweets in measuring the interim quality of the associated research papers. Tweets of 53,831 computer science papers from the Microsoft Academic Graph (MAG) dataset were extracted for sentiment classification. The non-neutral sentiment keywords and the attributed aspects of the papers were manually identified. Findings indicate that although neutral sentiments are majorly found in tweets, the research impact of papers which had all three sentiments was better than papers which had only neutral sentiment, in terms of both bibliometrics and altmetrics. Implications for future studies are also discussed.