Session: Computational Approaches to Social Work Research Utilizing Text As 'big Data' (Society for Social Work and Research 29th Annual Conference)

Social work is increasingly navigating the possibilities and perils of 'big data' and artificial intelligence (AI) in terms of scholarly contributions and shaping social policy. One important contribution the field is making is innovating the ways in which text, language, speech, and other qualitative data sources are searched and extracted for classification and algorithmic modeling using AI. By coupling expertise in qualitative research, guided by principles of social equity, social good, and anti-racism, with the analytic power of computational approaches, social work is advancing the field of data science for maximum public impact. In this symposium, we present four use cases of utilizing text - and language more broadly - as 'big data' in data science and social equity research.

The first paper presents a study that uses outpatient behavioral healthcare clinic notes as 'big data' and natural language processing methods to evaluate note quality. Based on human expertise, an algorithm was developed to capture note readability, clinical content, and person-centeredness. This study applied the algorithm to compare the quality of clinic notes generated by AI to those written by clinicians. AI-generated notes were longer, used more jargon, and included less individualized information compared to clinician-written notes. This study demonstrates the potential utility of AI tools in evaluating note quality and the importance of anchoring these tools in human clinical expertise.

The second paper presents social media posts and metadata to detect and classify anti-Asian hate speech and predict racist and allyship sentiments. Researchers analyzed tweets and developed a codebook to identify racist and pro-Asian sentiments. This study further employed large language models for classification tasks using zero shot and few shot learning. While ChatGPT showed promise in detecting explicit hate speech, it was less accurate in detecting subtle forms of discrimination. The study highlights the complexity of this issue and the need for more nuanced detection methods.

The third paper presents web-scraped text data on clinical social worker demographics to identify gaps in services for racial, ethnic, and linguistic minorities seeking mental health care in the United States. This research examines the diversity of the US social work profession using web scraping and machine learning techniques. Findings reveal that social worker licensure is associated with a lower probability of identifying as Asian or Black, and no significant association with identifying as Hispanic or Latino. The study suggests potential gaps in the licensing and training system that contribute to these disparities and hinder efforts to create a more diverse and culturally competent mental health workforce.

Finally, the fourth paper uses LENA technology to record and analyze speech and language data from home-based interactions and code caregiving patterns to assess father-child interactions. Early verbal engagement is associated with healthy psychosocial and cognitive development in young children, however, few studies exist on the role of fathers from low-income urban families using observational data. This study seeks to fill this gap and demonstrate the utility of a novel technology to assess early parent verbal communication for descriptive and intervention research purposes.