Sentiment analysis gets increasing attention in software engineering with new tools emerging from new insights provided by researchers. Existing use cases and tools are meant to be used for textual communication such as comments on collaborative version control systems. While this can already provide useful feedback for development teams, a lot of communication takes place in meetings and is not suited for present tool designs and concepts. In this paper, we present a concept that is capable of processing live meeting audio and classifying transcribed statements into sentiment polarity classes. We combine the latest advances in open source speech recognition with previous research in sentiment analysis. We tested our approach on a student software project meeting to gain proof of concept, showing moderate agreement between the classifications of our tool and a human observer on the meeting audio. Despite the preliminary character of our study, we see promising results motivating future research in sentiment analysis on meetings. For example, the polarity classification can be extended to detect destructive behaviour that can endanger project success.
翻译:现有的使用案例和工具意在用于文字交流,例如关于合作版本控制系统的评论。虽然这已经可以为发展团队提供有用的反馈,但许多沟通是在会议上进行的,不适合目前的工具设计和概念。在本文中,我们提出了一个能够处理现场会议音频和将转录的语句分类为情感极极化类的概念。我们把公开源语音识别的最新进展与先前的情感分析研究结合起来。我们测试了学生软件项目会议的方法,以获得概念证明,显示我们工具分类与会议音频上的人类观察员之间适度一致。尽管我们的研究具有初步性质,但我们看到在会议情感分析方面的未来研究有希望的结果。例如,极化分类可以扩大,以探测可能危及项目成功的破坏性行为。