Intelligent question generation method and device and computer-readable storage medium

Abstract

The invention discloses an intelligent question generation method and device and a computer-readable storage medium, which are used for automatically generating questions for an input article, and outputting the same. The method comprises the steps of: S1, utilizing a seq2seq model for extraction of key contents on the article; S2, carrying out syntactic analysis and named-entity identification oneach sentence in the key contents to establish a corresponding grammar tree of each sentence; S3, utilizing the grammar tree for matching with question templates in a pre-established question template database, and if a matched question template exists, converting a sentence, which corresponds to a grammar tree, into a questioning sentence based on the matched question template, and thus generating a question; and S4, utilizing a neural network to sort the generated questions, and then outputting the same.

Description

technical field [0001] The present invention relates to the technical field of computer and natural language processing, and in particular to a method for generating intelligent questions based on deep learning and a corresponding device. Background technique [0002] With the vigorous development of computer networks, there are more and more informatized data on the network, and it is impossible for users to find their own points of interest after reading all the information. At present, most of the screening and reading of articles and documents on the Internet is still based on browsing the titles to decide whether to continue reading the articles. However, the defect of this method is that many titles cannot accurately and comprehensively reflect the core content of the article. , so that the user cannot accurately find the article of interest, or, miss the article of interest. Therefore, we can consider using related technologies of natural language processing to conde...

AI Technical Summary

Problems solved by technology

At present, most of the screening and reading of articles and documents on the Internet is still based on browsing the titles to decide whether to continue reading the articles. However, the defect of this method is that many titles cannot accurately and comprehensively reflect the core content of the article. , resulting in users being unable to accurately find articles of interest, or missing articles of interest

Kunichika et al. used grammatical relationships to generate questions about English stories to test different people's understanding of stories. In this paper, they generated questions from five angles: asking questions about the content of the entire sentence, using the correspondence between synonyms and antonyms, and asking questions based on time and space. Questions about the relationship between sentences, words with plural forms in sentences, and plural related phrases, but the results show that there are more grammatical errors in the questions generated in this way

[0012] However, the above methods still have obvious deficiencies in dealing with complex Chinese: because Chinese articles with complex structures and implicit expressions through layer-by-layer rendering are not as straightforward as educational articles, lengthy texts are generated in the above prototype system The problem does not meet our needs. For example, the repetition of similar problems directly reduces the satisfaction of generating problems, and what we want are only those problems that locate the key content in the text, so lengthy text is not a good problem for problem generation. input; in addition, using the generation order of the questions as the final output order of the questions, or using a simple sorting method such as linear regression to sort and then output the questions, cannot ensure that each question is ranked in the most appropriate position

Images