Health Topics Mining in Online Medical Community

With the development of medical informatization, more and more patients actively obtain health information from online medical communities. The traditional methods based on statistical analysis are inefficient in dealing with growing mass of medical texts. Based on the Latent Dirichlet Allocation (LDA), we propose the Medical of Sentence LDA (MS-LDA) for short online medical texts with distribution features of medical words in online medical communities. Disease-related hot topics are assumed to be generated by sentences, the Gaussian function is employed to fit word distribution, and the correlation weight is exploited to modify word frequency for the information extension in sentences. Furthermore, Unified Medical Language System (UMLS) is introduced to cluster the topic recognition results from disease-related hot topics. Experiments on three representative disease boards from www.MedHelp.org show that the perplexity value and word relevance in topics are significantly improved by MS-LDA. Besides, hot topics concerned by members are automatically mined and texts in online medical community are automatically classified.