Multi-label text classification method and system based on dynamic weight contrastive learning

Abstract

The invention relates to the technical field of information detection, and proposes a multi-label text classification method and system based on dynamic weight comparison learning, wherein the method includes: preprocessing to obtain training set word vectors and label sequences; The classification model of the long-term memory network encodes the multi-label text of the training set to obtain a vectorized representation containing semantic information; decodes the label sequence and the vectorized representation containing semantic information to obtain the predicted label sequence; through the joint cross-entropy probability distribution Loss and Contrastive Learning Loss Calculate the loss between the predicted label sequence and the label sequence, and optimize the classification model as a multi-label text classification model according to the loss; input the multi-label text of the test set to be classified into the multi-label text classification model, and the output corresponds to the final tag sequence. According to this method, the multi-label semantic confusion phenomenon and the long-tail problem in multi-label classification datasets are effectively solved.

Description

technical field [0001] The invention relates to the technical field of information detection, in particular to a method, system, electronic device and computer-readable storage medium for multi-label text classification based on dynamic weight comparison learning. Background technique [0002] Multi-label classification tasks have a wide range of application scenarios in the NLP field, including but not limited to text classification, label recommendation, and information retrieval. Multi-label classification scenarios are different from multi-classification scenarios: in multi-classification scenarios, a piece of data has only one label, but this label may have multiple categories. For example, in sentiment classification tasks, judging the sentimental tendency of a text is usually classified as One of "positive", "negative" or "neutral"; while in a multi-label classification scenario, a piece of data may have multiple labels, for example, a piece of news may be classified ...

AI Technical Summary

Problems solved by technology

[0003] Existing multi-label text classification methods generally directly predict the probability of text corresponding to each label when dealing with multi-label classification tasks. This method ignores the correlation between labels, and has poor prediction performance for low-frequency labels, that is, there is a long tail problem.

Images