Yang F.-C.Lee A.J.T.Kuo S.-C.2019-07-242019-07-24201601485598https://scholars.lib.ntu.edu.tw/handle/123456789/415108With the rapid development of the Internet, more and more users utilize health communities (known as forums) to find health-related information, share their medical stories and experiences, or interact with other people in the communities. In this paper, we propose a framework to analyze the user-generated contents in a health community. The proposed framework contains three phases. First, we extract medical terms, including conditions, symptoms, treatments, effectiveness and side effects to form a virtual document for each question in the community. Next, we modify Latent Dirichlet Allocation (LDA) by adding a weighted scheme, called conLDA, to cluster virtual documents with similar medical term distributions into a conditional topic (C-topic). Finally, we analyze the clustered C-topics by sentiment polarities, and physiological and psychological sentiment. The experiment results show that conLDA outperforms the original LDA, and can cluster relevant medical terms and relevant questions together. The C-topics clustered by conLDA are more thematic than those clustered by the original LDA. The results of sentiment analysis may provide a quick reference and valuable insights for patients, caregivers and doctors. ? 2016, Springer Science+Business Media New York.Health social mediaLatent Dirichlet AllocationSentiment analysis[SDGs]SDG3caregiver; experimental model; extract; human; human experiment; mining; social media; symptom; consumer health information; data mining; procedures; social media; statistics and numerical data; Consumer Health Information; Data Mining; Humans; Social Mediajournal article10.1007/s10916-016-0604-42-s2.0-84988649896https://www.scopus.com/inward/record.uri?eid=2-s2.0-84988649896&doi=10.1007%2fs10916-016-0604-4&partnerID=40&md5=8525b27d62ac9ce19d12870801801fd5