Text abstract generation system, method, device and computer readable storage medium

Abstract

The invention provides a text abstract generation system, a method, a device and a computer readable storage medium, and aims at forming training data through a seq2seq framework and text-abstract pairs to realize a system for generating abstracts for texts. A classic seq2seq framework is used for solving the problem from an uncertain length sequence to a sequence, a Bert Chinese pre-training model is used at an Encoder end to obtain text input information, a two-way decoding mechanism is introduced at a Decoder end, and the quality of a generated text is improved to a certain extent. When theabstract sequence is finally generated, top K results in the two directions are cached at the same time according to the beam-search thought, and finally the result with the highest probability is found and serves as output.

Description

【Technical field】 [0001] The present invention relates to the technical field of computer word processing, and in particular to a text abstract generation system, method, device and computer-readable storage medium based on a sequence-to-sequence framework of a bidirectional decoding mechanism. 【Background technique】 [0002] Text summarization refers to refining and summarizing the content, and summarizes the main content that users are concerned about with a concise and intuitive summary, which is convenient for users to quickly understand and browse massive content. [0003] With the advent of the Internet age, text summarization is becoming more and more important. First, there are more and more information explosion texts, and how to receive more information in a limited time has become a thorny issue; second, there is too much relevant information, redundant, one-sided, and impurity information leads to information overload; Finally, the popularization and use of mobi...

AI Technical Summary

Problems solved by technology

However, there is still a problem with the generative summary, that is, the quality of the summary sequence will become worse and worse, because the word probability generated earlier will be high, but as the sequence prediction deepens, the conditional probability will decrease, so that the generated word probability will also decrease. The resulting phenomenon is that the generation accuracy of the first few words will be higher than that of the last few words. If the sequence is generated from right to left, the generation accuracy of the next few words will be higher than that of the first few words, that is, no matter from which Direction generation will have the problem of directional tilt

Images