Disease Topic Modeling of Users' Inquiry Texts: A Text Mining-Based PQDR-LDA Model for Analyzing the Online Medical Records

Disease information mining is one of the critical factors affecting users' perception of the disease and has attracted extensive attention from the information management community in recent years. If the mined disease information is incompatible with the disease information perceived by the user, it will eventually lead to the loss of users from the online medical consultation platform, degrading its operation and management. Using existing models to mine disease information leads to significant errors when users perceive the disease. Therefore, this research extends the latent Dirichlet allocation (LDA) and Twitter-LDA models to propose an intelligent topic model, PQDR-LDA. Compared with the Twitter-LDA model, the proposed model has a smaller perplexity value, stronger generalization ability, greater coherence value, lower correlation between topics, and stronger ability in extracting the disease information. It is found that the accuracy of disease diagnosis is very low, and the user's need for perceiving the disease will be reduced while using the traditional model to mine only the text of user questions on an online medical consultation platform. The accuracy of disease diagnosis does not decrease while only mining the doctor's reply text. Disease information that is more suitable for the consultation text can be obtained, which in fact cannot meet the user's real appeal for health, and reduces the users’ needs in perceiving the disease. These findings have important management implications for the platform's operation and decision-making. Besides, users will ask questions in more medical texts simultaneously, which makes things more complicated. Unique management insights are obtained based on the disease information mining of user consultation texts through multiple consultation texts and multiple doctor replies.