Procedures for training a ChatGPT-based classification for mental health screening in a college population

Abstract

Providing mental health remains one of the greatest challenges in world health care. For proper screening standardized instruments are unavoidable, but expert clinical interviews are even more critical. However, this remains inaccessible to large human groups. Recent developments in artificial intelligence [1] can help to help close this gap, especially for college students for it is easier to access their needs, language and mores, and because young people tend to be less reluctant to interact with chat Apps in their mobile devices. What is lacking, and however offered here, is a the-oretically grounded framework for dealing with the relationship between human spontaneous re-sponses (explanations to personal challenges which include their cognitions and feelings) to ap-propriate, expert-guided inquiries, and the potentially diagnostic information contained in these utterances. Following Weiner's [2] 3-dimensional approach to extracting causal attributions from natural language (which have been shown in the scientific literature to be associated with depression and anxiety [3], optimism, and self-efficacy), here we describe procedures we used for 1) collecting these attributions in a sample of college alumni from a large university in central Mexico, 2) recog-nition of sociocognitive meaningful patterns on these utterances [4], 3) to classify them (with the help of mental health experts) into 10 fuzzy sets (8+2) and to estimate their associated statistical and computational criteria, and 4) to check the diagnostic potential of these cut-points and classifications for rapid automated mental health screening. We show some exemplars of the iterative process using chatGPT-3 to obtain stable and acceptable network parameters. The ultimate aim is to validate the model and apply this classifier on-the-fly (in a voluntary and automated way through chatGTP-4), during the admission process of about 40,000 freshmen by 2023, in order to propose and provide a theoretically grounded, and almost personalized, institutional support in mental health prevention.

Ähnliche Arbeiten