Keyword extraction method and system for financial and economic messages

Abstract

The invention discloses a keyword extraction method and system, and the method comprises the following steps: obtaining financial short message text data, and marking a financial text; inputting the labeled text data into a pre-trained convolutional neural network to obtain font embedding feature vectors of text data characters; inputting the labeled text data into a pre-trained RoBerta-wwm model,and obtaining a semantic embedding feature vector of a text data character; carrying out splicing and dimensionality reduction on the font embedded feature vector and the semantic embedded feature vector, and obtaining a combined character feature vector; inputting the combined character feature vector into a conditional random field layer, and obtaining an output character label by adjusting training parameters; and performing extraction of keywords from character tags. According to the method and system, a Chinese RoBerta-wwm prediction model is adopted to represent character vectors of financial short message texts, representation is carried out in combination with five-stroke characteristics of Chinese, and the extraction accuracy of keywords can be improved in combination with five-stroke font characteristics of the Chinese.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a method and system for extracting keywords from financial newsletters. Background technique [0002] At present, most text keyword extraction algorithms are based on unsupervised algorithms. Existing keyword extraction methods include: keyword extraction methods based on statistical features, keyword extraction methods based on word graph features, and keyword extraction methods based on topic models And the combination of the above keyword extraction methods. However, the existing keyword extraction methods rely heavily on the performance of the Chinese word segmenter, and the Chinese word segmenter has a high proportion of misclassification of proper nouns in the financial field, and the extracted keywords are not accurate. For For short texts such as financial newsletters or even ultra-short texts of more than a dozen characters, the text statistical features, word map ...

AI Technical Summary

Problems solved by technology

[0002] At present, most text keyword extraction algorithms are based on unsupervised algorithms. Existing keyword extraction methods include: keyword extraction methods based on statistical features, keyword extraction methods based on word graph features, and keyword extraction methods based on topic models And the combination of the above keyword extraction methods. However, the existing keyword extraction methods rely heavily on the performance of the Chinese word segmenter, and the Chinese word segmenter has a high proportion of misclassification of proper nouns in the financial field, and the extracted keywords are not accurate. For For short texts such as financial newsletters or even ultra-short texts of more than a dozen characters, the text statistical features, word map features, and topic features used by existing solutions are relatively weak, and the keywords extracted by using existing solutions cannot effectively express financial newsletters The core purpose of the keyword algorithm leads to a low recall rate of the keyword algorithm

Images