Text classification model training method, and text classification method and device

Abstract

The invention discloses a text classification model training method, a text classification method and a text classification device. The text classification model training method comprises the steps ofscreening out a plurality of feature words by computing word frequency-inverse document frequency and mutual information of each candidate word in a training text set; and training a text classification model according to a genetic algorithm based on each feature word. According to the technical scheme of the method provided by the embodiment of the invention, the plurality of feature words are screened out by computing word frequency-inverse document frequency and mutual information of each candidate word in the training text set, and thus the problem that in the prior art, the screened feature words are low in accuracy can be solved.

Description

technical field [0001] The present disclosure generally relates to the field of computer technology, and in particular relates to a text classification model training method, a text classification method and a device thereof. Background technique [0002] After the product is put on the market, users can give feedback on comments or problems in the process of using the product through text, which is the embodiment of user experience. Therefore, user feedback has a high reference value for improving the user experience of the product. However, there are tens of thousands of user feedback problems every day. How to dig out effective information from a large amount of feedback data, discover product problem improvement points, and conduct timely monitoring and alarm for product sudden problems, and effectively prevent and control public opinion. appears to be particularly important. How to classify the text information of user feedback becomes the basis for processing a large ...

AI Technical Summary

Problems solved by technology

[0007] In the feature extraction stage of text classification, although the traditional mutual information feature extraction method has the characteristics of enhancing the connectivity between categories and features, the effect on feature extraction with scattered distribution is not good. low accuracy

[0008] In the training phase of the classification model, most of the existing methods use manual construction of the classification model, resulting in low classification accuracy.

[0009] In the classification stage, what the existing technology actually solves is the tree-like classification between multiple categories in a single layer, and does not solve the classification problem of tree-like multi-layer categories in practical applications

Images