Imbalanced text classification method introducing keyword features

Abstract

The invention provides an unbalanced text classification method introducing keyword features, which comprises the following steps of: firstly, defining a hierarchical classification system aiming at the field of military news, including 32 leaf categories; extracting keywords of each category by utilizing normalized point mutual information and improved information gain; and carrying out training by fusing the keyword features and the neural network semantic features. Through the steps, for the unbalanced text classification problem, class label distribution serves as prior information to be introduced into a text classification model, and text content and class keyword information are utilized in the training process; performing category keyword selection by using normalized point mutual information and improved information gain as statistical magnitude of category keyword selection; the keyword information and the text information are utilized to jointly train the text classification model, and the problem of class imbalance in text classification is effectively solved.

Description

technical field [0001] The invention relates to a text classification method, in particular to an unbalanced text classification method that introduces keyword features. Background technique [0002] Category imbalance means that there is a long tail phenomenon in the corpus, and the amount of data in each category varies greatly. The existing text classification models are essentially the optimization of the loss function, which minimizes a certain loss function. In imbalanced text classification, because the sample size of each category is different, the large-scale sample loss function will account for a high proportion, and eventually the model will be biased towards the large-scale category. [0003] For the common text classification problem, the existing solution is to obtain the vector representation of the text by pre-training the language model bert, and then use the cross-entropy loss function for fine-tuning. For the problem of class imbalance, it is mainly imp...

AI Technical Summary

Problems solved by technology

However, neither of these two methods can solve the underfitting problem of sparse categories. Suppose a sparse category has only one sample, but the sample contains more words, and the combination of these words will only appear in this category.

Neither resampling nor class weighting can address the diversity of this wash sample

Images