Keyword extraction method and device, encoder and decoder

Abstract

The invention discloses a keyword extraction method and device, an encoder and a decoder, and the method comprises the steps: transmitting a keyword generation process, namely decoding information, to an encoding processing part in a manner of dynamically modifying a syntactic graph structure of an original document through the generated keyword information; therefore, the two problems of modeling the keyword relationship and obtaining better semantic representation are solved at the same time, and the document coding effect is improved. Furthermore, more accurate and diversified keyword sequences are obtained by matching with a diversity inference process. In other words, the keyword generation accuracy is improved, the generated repeated keywords are reduced, and the total number of the generated keywords is increased.

Description

technical field [0001] The present application relates to, but is not limited to, information processing technology, especially a keyword extraction method and apparatus, encoder and decoder. Background technique [0002] Keyword extraction is to extract the core content to be expressed in the text by some means given a long text, so as to accurately and quickly extract key information from a large amount of information. These keywords can be entities with specific meanings, or some basic concepts or events. The extracted keywords can be represented by a keyword sequence, and the keyword sequence can be listed in order according to the degree of confidence. The higher the degree of confidence, the higher the ranking. The extracted keyword sequences can be applied to the subject tags of articles in different fields such as travel notes, notes, news, etc., as well as document retrieval and recommendation systems. [0003] In the related art, one way is to extract keywords by...

AI Technical Summary

Problems solved by technology

This approach relies too much on human-written rules, and the overall effect is not good enough

[0004] Another way is to extract keywords based on sequence-to-sequence model, however, this way faces two problems: 1) How to generate a document representation good enough to reflect the most important key words in the original document for keyword extraction Semantic information; 2) How to model the relationship between keywords in the keyword set, that is, how to better learn the conditional probability P(y n |y ), where y n is the keyword that needs to be generated currently, y is the generated keyword sequence

Images