Stack Overflow (SO) is a widely recognized online question-and-answer platform for programming, which has also fostered a substantial community dedicated to machine learning (ML), providing a space for both novices and experts to exchange ideas and find solutions to ML-related problems. However, as a relative minority of this online programming platform, research has demonstrated lower engagement in the ML community, but it remains largely unexplored to understand what hinders the engagement and contribution from ML users' perspectives. This paper presents an empirical study based on 22 hours of semi-structured interviews and 131 survey responses with users on SO and reveals the key factors that may lead to the lower response rate and extended waiting time for ML questions on SO, which includes the unique quality requirement for posting ML questions, the discrepancy between time invested and benefits gained, the dispersed nature of the ML community across various platforms and the desired improvement for SO. Moreover, the qualitative study reveals a declining friendliness in SO's culture over time; the subsequent quantitative study corroborates that newcomers frequently encounter stress when posting and answering ML questions, even though this stress diminishes with increased experience. Additionally, we also explored the potential influence of generative AI tools (e.g., ChatGPT) on online question-and-answer platforms, specifically focusing on ML Q&A. We hope the results of this study can pave the way for enhancing the experience of ML users on online platforms, ultimately facilitating improved knowledge exchange and collaboration within the ML domain.
Read full abstract