A method and system for generating machine reading comprehension models considering topics

Abstract

The invention discloses a method and system for generating a machine reading comprehension model considering topics. The invention extracts potential topic information in training sample data, and uses the topic information to supervise the training of the reading comprehension model, thereby improving reading comprehension The effect of the model. The model disclosed in the present invention considers extracting multiple topics corresponding to training samples before model training, and uses the topic information of the samples to improve the effect of machine reading comprehension. The basic process of this method is: process each training sample and find out the vector representation that can represent this sample; cluster the samples and find the mean value of similar sample vectors as the vector representation of the topic; use The attention mechanism gives more weight to the words whose vector representation is more similar to the topic vector of this sample. In addition, the training data can achieve better results after better data cleaning, because we can get better topic vector representation after noise reduction.

Description

technical field [0001] The present invention relates to the field of machine reading comprehension in the field of natural language processing, and more specifically, relates to a method and system for generating a machine reading comprehension model considering topics. Background technique [0002] Reading comprehension is the understanding based on reading, which can be abstractly summarized as the process of extracting information from text and understanding meaning through reading. It is a very routine test content in our traditional language disciplines. The general form is to give you an article, and then ask some questions about these articles. Students answer these questions to prove that they understand the main content of the article. , the closer the answer is to the standard answer, the more thorough the understanding of the article is. [0003] Machine Reading Comprehension (Machine Reading Comprehension), as the name suggests, is to let machines replace humans...

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the present invention is that, when performing machine reading comprehension in the prior art, the machine does not actually understand the meaning of the words and sentences in the paragraph, but only knows which word is more likely to be the beginning word and the sentence of the answer through training. Sentence ending words, which words are more relevant to the technical defects of the question, provide a machine reading comprehension model generation method and system that considers the topic

Images