Abnormal event classification method and system based on text processing

Abstract

The invention discloses an abnormal event classification method and system based on text processing, and the method comprises the steps: carrying out the classification of a text through naive Bayes, and proving that the text is highly correlated with a label; performing abnormal event label classification by utilizing a BERT model; judging according to the combination of a BERT model and a service logic model, adjusting the output weight of the BERT according to an artificial rule, learning joint probability distribution from input to output by taking the independence among naive Bayesian model feature words as a premise hypothesis, and solving an output enabling the posterior probability to be maximum under an input condition according to the probability distribution; and in combination with naive Bayes, a BERT model and a service logic model, automatic classification and labeling of abnormal event data are realized, so that an abnormal event analysis service of the security industry is assisted. The BERT model is used for abnormal event classification and is combined with the naive Bayesian model, related business knowledge is fused for different types of label classification tasks, and different model adjustments are carried out.

Description

technical field [0001] The invention relates to a natural language processing technology, in particular to a method and system for classifying abnormal events based on text processing. Background technique [0002] Abnormal event classification refers to the label classification and marking of an abnormal event through natural language processing algorithms. The tags include alarm type, event type, case type, and address area type. The current text classification of abnormal events, the main algorithms are Naive Bayesian and TextCNN. [0003] Among them, Naive Bayes is a simple model that relies on the bag-of-words model. It can only count macroscopic semantic features, and cannot obtain word position information and the relationship between words. It is a low-level model with low accuracy. TextCNN, on the other hand, is a convolutional neural network that uses text as input. It has the following main defects: 1. It relies on word segmentation and trained word vectors. For ...

AI Technical Summary

Problems solved by technology

[0003] Among them, Naive Bayes is a simple model that relies on the bag-of-words model. It can only count macroscopic semantic features, and cannot obtain word position information and the relationship between words. It is a low-level model with low accuracy. TextCNN, on the other hand, is a convolutional neural network that uses text as input. It has the following main defects: 1. It relies on word segmentation and trained word vectors. For abnormal event texts, training a set of general word vectors needs to rely on place names, The identification of entities such as proper nouns requires a lot of work; 2. The classification of an abnormal event can only be reflected by a few words in the text, and the neural network cannot focus on specific segments of the input text sequence, resulting in The discrimination accuracy is low; 3. The convolutional neural network uses a sliding window to scan the input text, so the connection between non-adjacent words cannot be obtained, and it is difficult for the model to obtain a better semantic representation ability

Images