Text multi-label analysis method, device, electronic equipment and storage medium

Abstract

The invention discloses a text multi-label analysis method, device, electronic equipment and storage medium. The method includes: acquiring training text data, the training text data includes multiple texts, and performing word segmentation on the training text data; using N-gram Word segmentation is modeled to form a text collection that includes a plurality of word sequence sets, as a corpus, the word sequence set includes a plurality of word sequences made up of N words; the LDA model is trained using the corpus, and the The training results are scored and verified, and the LDA model with high score is selected for persistence to obtain the persistent LDA model; the persistent LDA model is used to extract the topic cluster of the test text data, and the label is defined in combination with the semi-supervised method of the term to determine the Cluster labels for topic clusters. The invention combines LDA and N-gram, considers the sequence of words, and has the characteristics of LDA to quickly generate topic clusters, can accurately obtain text topic clusters, and obtain text topic meanings.

Description

technical field [0001] The invention relates to the field of data processing, in particular to a text multi-label analysis method, device, electronic equipment and storage medium. Background technique [0002] Automatically define the subject meaning of a text. The current method is mainly keyword definition or text classification. Among them, the keyword definition methods are mostly extraction and generation. The extraction method directly intercepts the words representing the topic of the article from the text. , and the generative model is better than some articles that lack key expressions. However, due to the complexity of the Chinese context, such as the comment text "These can only be explained or not enough", the current mainstream model generative architecture seq2seq is too implicit for these Regardless of the lack of understanding of the short text contained in it, the response performance of the service cannot meet the rapidly increasing data volume processing n...

AI Technical Summary

Problems solved by technology

However, since the essence is also a bag-of-words model, the disadvantage is that if the word segmentation is not accurate and the regional term changes have a substantial impact on the results, the final modeled subject term correspondence matrix is ​​difficult to meet the precise requirements of Chinese semantics

[0005] For the above problems, there is no better solution

Images