Interpretability analysis method based on CNN text classification model

Abstract

The invention discloses an interpretability analysis method based on a CNN text classification model, and the method comprises the steps: obtaining one or more pieces of original text data, and carrying out the preprocessing of the original text data; constructing a text classification model based on a convolutional neural network, converting the preprocessed original text data into a distributedmatrix by using the text classification model, and performing classification prediction based on the distributed matrix to obtain a text classification result; backtracking and analyzing the importance of each identifier influencing the text classification result in each dimension to generate an importance vector matrix; and generating a visual analysis graph based on the importance vector matrix.According to the method, the contribution value of each dimension of each identifier in the text to the prediction result can be quantitatively determined on the basis of the interpretability of theCNN text classification model and the generation reason of the reverse backtracking text classification result, and the analysis result is presented through the visual graph.

Description

technical field [0001] The present invention relates to the technical field of text classification, and more specifically relates to an interpretability analysis method based on a CNN text classification model. Background technique [0002] Convolutional neural networks are built to imitate biological visual perception mechanisms. Initially, they have made great progress in the field of computer vision, and in recent years they have gradually developed rapidly in the field of natural language processing. There is a very important application in the field of natural language processing, which is text classification, which is based on the content of the text, to predict the category of the text, such as: judging whether the emotional tendency of a comment is praise or criticism (two classifications), judging a paragraph News is a category of finance, sports, education, politics, etc. (multi-category). If the text belongs to only one category, it is called a single label. If it...

AI Technical Summary

Problems solved by technology

[0004] At present, there are relatively few studies on the interpretability of text classification models, especially those based on CNN models, and they have certain limitations. degree of visualization

Images