Abstract: 【 Purpose/significance】 In recent years, online health communities for the elderly have become an important medium for the elderly to share pain and information support. To ensure that the user group can effectively obtain the health information they need, this article proposes a user health information recommendation model based on the elderly online health community. 【Method/pro⁃ cess】 Firstly, construct a heterogeneous information network based on users, topics, and their emotional tendencies. Secondly, from the perspectives of content and network structure, support vector machine models are used to classify topics and user emotional roles. Fi⁃ nally, based on the user's emotional attributes, topic emotional attributes, and topic similarity calculation results, the user's informa⁃ tion needs are predicted.【 Result/conclusion】 Research shows that users tend to post positive posts, but tend to comment on negative content in others' posts. Compared to neutral content expression without obvious emotional tendencies, user interaction with emotional tendencies is more common in communities than information support.【 Innovation/limitation】 The user health information recommen⁃ dation model proposed in this article helps to attract and encourage more elderly people to share and obtain health information. The model in this article has a certain contribution to the application of machine learning in mental health. But in future work, interview and survey methods should be included to enhance the theoretical depth of the recommendation model.