Recognition of Hotspot Words for Disease Symptoms Incorporating Contextual Weight and Co-Occurrence Degree

Identifying hotspot words associated with disease symptoms is paramount for disease prevention and diagnosis. In this study, we propose a novel method for hotspot word recognition in disease symptoms, integrating contextual weights and co-occurrence information. First, we establish the MDERank model, which incorporates contextual weights. This model identifies words that align well with comprehensive weights, forming a collection of disease symptom words. Next, we construct a graph network for disease symptom words within each time period. Utilizing the graph attention network model, we incorporate word co-occurrence degree to identify potential hotspot words associated with disease symptoms. We conducted experiments using user-generated posts from the Dingxiangyuan Forum as our data source. The results demonstrate that our proposed method significantly improves the extraction quality of disease symptom words compared to other existing methods. Furthermore, the performance of our constructed recognition model for disease symptom hotspot words surpasses that of alternative models.