Extraction type machine reading understanding method integrated with multiple paragraph information

Abstract

The invention provides an extraction type machine reading understanding method integrated with multiple paragraph information, and belongs to the technical field of reading understanding in natural language processing. A reading understanding system depended on the extraction type machine reading understanding method integrating the multiple pieces of paragraph information comprises a paragraph scoring device, a paragraph reader and an answer selector. The method comprises the following steps: S1, the paragraph scoring device obtains the possibility that paragraphs contain correct answers according to the correlation degree of questions and paragraphs; s2, the paragraph reader extracts the most possible N answers from the paragraphs according to the questions, and quantifies the possibility that the N answers are correct answers; and S3, results of the paragraph reader and the paragraph scoring device are fused by an answer selector, and thus multiplying the answer by the possibility of the paragraph where the answer is located to obtain the most possible answer in the whole article. The method gets rid of the limitation of the input length in the prior art, and can help the user to read and understand on the scientific research paper, so that the content of the paper is quickly known, and the latest progress in the field is closely followed.

Description

technical field [0001] The invention relates to an extractive machine reading comprehension method incorporating multiple paragraph information, and belongs to the technical field of reading comprehension in natural language processing. Background technique [0002] Machine reading comprehension is a technique that enables computer systems to understand the semantics of an input text and answer related questions. Because the task of reading comprehension can properly evaluate the computer system's ability to understand natural language, it has always been a research hotspot in the field of natural language processing technology. With the introduction of large-scale data sets, it is possible to train deep neural networks. For example, most mainstream machine reading comprehension methods use the SQUAD dataset proposed by Stanford University in 2016 for training and evaluation. [0003] On the other hand, since entering the 21st century, human science and technology have bee...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the technical defects that the existing machine reading comprehension model cannot effectively process long scientific papers due to the limitation of the input length, and the semantic similarity performance of the generated text needs to be improved even if the input length requirement is met. An extractive machine reading comprehension method for multiple paragraphs of information, the method can automatically read scientific and technological papers, and answer questions such as "what is the motivation of this paper (Motivation)", "what is the model like (Model)", "what is the experimental result?" Questions such as "Experiment", "What conclusion did the researchers draw (Conclusion)", and finally integrate all the answers into a complete paper explanation to help researchers obtain literature summaries with high semantic similarity, so as to quickly understand the paper Content, keeping up with the latest developments in the field

Images