学术文献库

Social aspects in software development teams are of particular importance for a successful project closure. To analyze sentiments in software projects, there are several tools and approaches available. These tools analyze text-based communication based on the used words to predict whether they appear to be positive, negative, or neutral for the receiver of the message. In the research project ComContA, we investigate so-called sentiment analysis striving to analyze the content of text-based communication in development teams with regard to the statement's polarity. That is, we analyze whether the communication appears to be adequate (i.e., positive or neutral) or negative. In a workshop paper, we presented a tool called SEnti-Analyzer that allows to apply sentiment analysis to verbal communication in meetings of software projects. In this technical report, we present the extended functionalities of the SEnti-Analyzer by also allowing the analysis of text-based communication, we improve the prediction of the tool by including established sentiment analysis tools, and we evaluate the tool with respect to its accuracy. We evaluate the tool by comparing the prediction of the SEnti-Analyzer to pre-labeled established data sets used for sentiment analysis in software engineering and to perceptions of computer scientists. Our results indicate that in almost all cases at least two of the three votes coincide, but in only about half of the cases all three votes coincide. Our results raise the question of the "ultimate truth" of sentiment analysis outcomes: What do we want to predict with sentiment analysis tools? The pre-defined labels of established data sets? The perception of computer scientists? Or the perception of single computer scientists which appears to be the most meaningful objective?