Inferring Soft Skills from Developer Interactions: A Data-Driven Approach Using the TAWOS Dataset

Abstract

Soft skills like collaboration, communication, and proactivity are increasingly critical in the era of Artificial Intelligence, becoming even more important than technical expertise. However, identifying and assessing these human skills remains challenging, as they are often not documented, are based on personal opinions, and are primarily identified through human conversations. This study addresses this challenge by showing how developer comments in issues can help to understand soft skills. The approach applied includes a practical and exploratory methodology, mixing network analysis to map collaboration patterns with machine learning and natural language processing. The TAWOS dataset was used as the primary data source, which is a public collection of open-source projects, focusing on comments in issues and how developers interact. The methods applied involved network analysis to map collaboration patterns, community detection to understand team structures, first responder analysis to infer proactivity, stylometric analysis to characterize communication style, and sentiment analysis to evaluate communication tone. The main findings, derived through a combination of empirical validation and systematic analysis, demonstrate that specific soft skills and ways of working together can be clearly seen from discussions on issues. This research gives new and practical insights into how developers work by exploring how to find soft skills in software engineering. It helps to better understand how individuals and teams behave, supporting their professional growth and aiding project management in resource allocation.