A Case Study of Research Data Preparation Using ChatGPT

Abstract

　As an effective use of ChatGPT in research activities, the authors proposed a method to prepare research data from a set of relatively short sentences. The titles of 239 articles published in four journals were used as a case study. Prompts were given to ChatGPT to infer the content from each title, and the obtained responses were organized. Therefore, the authors reduced the effect of erroneous judgments by giving the same prompt ten times to ChatGPT and summing up the answers. As a result, when comparing two data sets generated by the same method, 97% of the data sets showed the same results, indicating high data reproducibility. However, depending on the slight difference in the wording of the prompt given to ChatGPT, the response obtained could also change. We need additional thought on the optimal wording of the prompts to get a correct response from ChatGPT.