Optimized Event Identification Fused with Entity Topic

Abstract

Event recognition model is very important for machine learning. In the existing event learning model, there are many kinds of models. For example, in the study of event recognition oriented to emergencies, it is mainly based on semantics and syntax, involving feature vectors and factors. For Internet events, there has also been progress from Boolean model, vector space model to probability model and language model. Aiming at the learning of news events, a more novel recognition model is expected to be developed and processed on the basis of the existing model, so that machine learning can have new ideas and more feature points to consider, the model will be more complete and the accuracy rate will be improved. Based on the existing topic model, we update the relationship between entity, entity topic, word topic and word. In the previous stage, we add the consideration of feature words, which we will mention later, and use concise graph theory to express. By comparing the predicted content with the original content and calculating the relationship with the threshold, the event recognition result and classification can be determined. The accuracy of the new topic recognition model is obviously improved compared with the original model.