Learning to Effectively Identify Reliable Content in Health Social Platforms with Large Language Models

With the widespread accessibility of the Internet, individuals can effortlessly access health-related information from online social platforms. However, the veracity of such health-related content is often questionable, posing a significant challenge in ensuring the content quality and reliability. The exponential growth in daily data generation necessitates the integration of artificial intelligence to assess the reliability of content shared on these platforms. In this paper, we focus on Large Language Models (LLMs) due to their outstanding performance. We introduce Health-BERT, a novel model built upon the BERT architecture. We fine-tuned Health-BERT using a carefully curated dataset from a prominent health information forum. Our experiments demonstrate the remarkable capabilities of our model, achieving an impressive accuracy rate of 94% even with relatively limited training data. This highlights the exceptional knowledge transfer capabilities of LLMs when applied to health-related content. Our model will be open-sourced, with the hope that this initiative will improve the identification of content reliability in health contexts.