Microblog sensitive event voice detection method based on unbalanced Bayesian classification

Abstract

The invention discloses a microblog sensitive event voice detection method based on unbalanced Bayesian classification. The method comprises the steps: S1, selecting a microblog voice data set needingto be detected, constructing an unbalanced data set through text feature processing, and constructing a classification model; assigning prior distribution of the classification model on the parameteromega, and randomly initializing the parameter omega to obtain an initial parameter vector omega0; S2, calculating an interval likelihood value of the classification model with the parameter of omega0 for each category of sub-data sets; S3, calculating the posterior probability of the classification model on the parameter omega 0; S4S4, sampling new parameter points; S5, recording a sampled parameter sequence; S6, the classification model calculates a probability distribution vector p that the to-be-tested voice features belong to each category, and predicts that the category of the to-be-tested voice features is the category with the highest probability in p; according to the method, a large number of data support training processes are not needed, and the problem of over-fitting of abnormal class samples is solved, so that the classification precision of the data set when the number of abnormal class voice is too small is effectively improved.

Description

technical field [0001] The invention relates to the technical field of data mining, in particular to a microblog sensitive event speech detection method based on unbalanced Bayesian classification. Background technique [0002] In the era of rapid Internet development, more and more people use the Internet to communicate, but the anonymity of the Internet itself will make people make irresponsible remarks on the Internet, including irresponsible comments on sensitive events, such as pornography Terrorism-related remarks, rumors, insulting remarks, etc. In social platforms such as Weibo, it is no longer feasible to manually screen Weibo speeches, and it is necessary to identify and detect these speeches through methods such as deep learning. However, in the task of sensitive event speech detection, most people’s microblogs do not involve sensitive events, and only a small number of people’s speeches involve sensitive events, resulting in a large difference in the number of n...

AI Technical Summary

Problems solved by technology

The function approximation method requires the training data set to reflect the real data distribution. However, in the detection of microblog sensitive event speech, there are few speeches related to sensitive events, and the data set has too few abnormal samples, which leads to the lack of abnormal samples. Unable to describe the real data distribution well, causing the model to overfit the abnormal samples

Images