Objective There are many violent topics in social networks. Violence topic identification is of great significance to the accurate intervention and control of network public opinion. At present， the research on cyber violence mainly focuses on the calculation of users'' negative emotions， violent user identification and other fields. There is a lack of fine-grained analysis and research on cyber violence， which can not accurately identify the attached carrier of cyber violence in the complex network environment.Method By analyzing the topic focusing characteristics of network violence， a topic recognition method of network violence based on LDA model and chi square test is proposed. Firstly， the topic in network corpus is recognized by LDA model， and the topic text is classified by similarity calculation method； Then the chi square test is used to screen the characteristics of violence in the topic text； Finally， the violence value of each topic is calculated according to the emotional dictionary， and the violence topic is judged according to the violence value.Result/Conclusion Experiments on a real network corpus verify the proposed method. The experimental results show that the average performance of violence topic recognition （F value） of this method is 80.64%， which is better than the comparison method， and achieves a good effect of network violence topic recognition.