Abstract: 【Purpose/significance】 Analyzing social media user data to detect their psychology and behavior has become a research frontier in the academic community. Health anxiety became one of the main psychological problems of the public. This paper aims to identify users with health anxiety tendencies in social media environments through automatic text classification.【 Method/process】 It used China's social media platform "Weibo" as the data source and identify users with health anxiety through data collecting, clean⁃ ing, tagging, text vectorization representation, and classification model construction.【 Result/conclusion】 This study found that the per⁃ formance of the RoBERTa-wwm model is superior to Bert-base-Chinese and other models. Combining the word vectors generated by the RoBERTa-wwm pre-trained language model with the psychological state feature vectors generated by the COMET model can bet⁃ ter represent the semantic features of text. The health anxiety detection model based on gated loop units, scaled dot product attention mechanism, and fully connected layer performs best. 【Innovation/limitation】 This study constructed various health anxiety detection models and evaluated their performances. It provided valuable references for constructing user health anxiety detection and early warn⁃ ing systems, as well as for the psychological safety monitoring of online health groups. Its limitation lays in the single data source.