Identifying Research Hotspots and Trends in Psoriasis Literature: Autotuned Topic Modeling with Agent.

Abstract

The rapid expansion of psoriasis research presents challenges in efficient analysis and trend identification, necessitating advanced approaches. We propose AgenTopic, an interactive topic modeling framework that integrates Bidirectional Encoder Representations from Transformers embeddings, dimensionality reduction, clustering, and a language model feedback loop to analyze the psoriasis research literature from 2000 to 2023. Applied to PubMed articles, AgenTopic extracted 158 psoriasis-related topics across 8 categories, outperforming traditional methods in handling complex medical literature. Further trend analysis using multiple modeling techniques, including a support vector regression-Linear model, revealed nonlinear patterns in research growth across categories (R² values = 0.75-0.97). Key trends identified include focus on nail psoriasis and spondyloarthritis, shift from TNF-α to IL-17 in pathogenesis understanding, rapid development of biologics and small-molecule inhibitors, and increased attention to comorbidities. We developed an interactive web tool to facilitate literature retrieval and identification of trends. To the best of our knowledge, this application of an agent-based interactive topic modeling framework to dermatological literature has not been previously reported. Using only topic-modeled data, our framework achieved a performance comparable with that of expert manual reviews in identifying research trends. AgenTopic performed better than several state-of-the-art topic modeling methods and demonstrated the potential of artificial intelligence for advancing medical literature analyses.