【Purpose/significance】The paper proposes a sentiment analysis method incorporating deep evolutionary features to improve the accuracy of Micro-blog sentiment analysis for public safety events.【Method/process】Taking the child abuse incident in Red, yellow and blue kindergarten as an example, the theme features and time features from Micro-blog are extracted by LDA and Crawler software, combining with the traditional text POS and emotion features. At last, the features are applied to XGBoost to generate a Micro-blog sentiment analysis integrated model.【Result/conclusion】The results show that the accuracy of emotion recognition is increased by 4% with the integration of evolutionary features, and the accuracy indicators of XGBoost classification is better than that of SVM and Random Forest. This proposed model can achieve better results in the sentiment analysis of public security events.