Evaluation of Data Sharing in Interaction-Based Emotion Recognition Research

Data is a fundamental part of all data-driven and artifi cial intelligence-related research. Independent verification of published results is crucial for ensuring research quality and promoting further advancements. However, reproducibility in data-focused research is a growing concern across disciplines. Our research aims to improve the reproducibility of interaction-based emotion recognition research. For this, we evaluate current data sharing practices in this area. We sur vey these practices by examining 100 publications published from 2014 to 2024. We examined the characteristics of shared and not shared data, trends over time, and influences of authorship and publication type. We requested data from the publications’ authors for which the data was not shared as part of the publication. Afterwards, we evaluated the rea sons we were given as to why the data could not be shared. Overall, we observed limited data sharing, with only 15 out of 100 publications pro viding their data, even after we sent requests for data to authors. We only received data for four publications after sending requests via email to the authors. Furthermore, the shared artifacts are often insufficient for full reproducibility. We found that, data was shared along a journal article more often than a conference publication. However, data shar ing increased noticeably in recent years, particularly since 2021. Overall, the lack of data sharing hinders the reproducibility and comparability of the research results. In the future, it is necessary to encourage the data sharing along publications by introducing mandatory guidelines.