Detecting Interpersonal Conflict in Issues and Code Review: Cross Pollinating Open- and Closed-Source Approaches

Interpersonal conflict in code review, such as toxic language or an unnecessary pushback, is associated with negative outcomes such as stress and turnover. Automatic detection is one approach to prevent and mitigate interpersonal conflict. Two recent automatic detection approaches were developed in different settings: a toxicity detector using text analytics for open source issue discussions and a pushback detector using logs-based metrics for corporate code reviews. This paper tests how the toxicity detector and the pushback detector can be generalized beyond their respective contexts and discussion types, and how the combination of the two can help improve interpersonal conflict detection. The results reveal connections between the two concepts. LAY ABSTRACT Software engineers often communicate with one another on platforms that support tasks like discussing bugs and inspecting each others‘ code. Such discussions sometimes contain interpersonal conflict, which can lead to stress and abandonment. In this paper, we investigate how to automatically detect interpersonal conflict, both by analyzing the text of the what the engineers are saying and by analyzing the properties of that text.