Academic Awards 2025 booklet

95 Sentiment in Software Engineering Unlike pervasive stereotypes, software engineering is not just a technical discipline; it's a profoundly human endeavor. In my thesis I studied how emotions expressed by developers in uence technical decision- making. Before my thesis it was already known developers convey feelings in code comments or bug reports, but the actual impact of these expressions on the development process was unclear, and this was the gap I studied. I combined a literature review, rigorous benchmarking of AI tools for sentiment detection, and a novel experiment. I found that developers express negative emotions to document and describe inePicient or problematic source code. Furthermore, I found how 30% of developers interpret negativity in these descriptions as a signal of priority. Thereby directly in uencing how technical issues are perceived and potentially addressed—even if unintentionally—despite many developers nding the usage of negativity as a proxy for priority problematic. Furthermore, I evaluated the AI tools commonly used to study emotions in software engineering, oPering crucial guidance to researchers by highlighting subtle but important diPerences in misclassi cations across diPerent tools. In summary, my thesis underscores how sentiment, even in descriptions of technical problems, aPects technical decisions further highlighting the importance of social dynamics in software engineering. Figure 1: A Directed Acyclic Graph (DAG) representing the causal model used to design the experiment that showed negativity aPects perceived priority. This model shows how we expect sentiment to in uence the perception of priority (operationalized as Priority, Urgency and EPort), and it was used to derive the statistical model t to analyze the data of the experiment. Figure 2: This gure shows Bayesian posterior plots illustrating whether software engineers are more likely to increase priority scores for negative descriptions of suboptimal code. This shows how, depending on the personal beliefs of the participants (Agree, No Opinion, Disagree), developers are more likely to increase the scores for the perceived EPort, Importance, and Urgency of suboptimal code if negativity is used to describe suboptimal code. Technical Debt Effort Experience Importance Manipulated Perception Sentiment Urgency