Implementation method and device for viewpoint type machine reading understanding

Abstract

The invention discloses an implementation method of viewpoint type machine reading understanding. The method includes the steps of: performing information extraction on a fusion result of each contextword vector; obtaining context information extracted based on the problem; and performing dynamic routing iteration by taking a contextualized coding result of the candidate answers as a clustering center and taking the context information extracted based on the question as clustering information, outputting the candidate answers weighted based on the context information, and identifying the optimal candidate answer from the candidate answers weighted based on the context information. According to the method, the optimal answer is matched for the dynamic selection type question, and the accuracy is high.

Description

technical field [0001] The present invention relates to the field of machine comprehension, and in particular, relates to a realization method of viewpoint-type machine reading comprehension. Background technique [0002] In machine comprehension, it is sometimes necessary to reason with contextual information in order to obtain correct comprehension results. For example, is it raining today? Is it better to eat Xiaochaihu or Ganmaoling Granules for a cold? Such questions need to be reasoned with contextual information in order to get an answer. [0003] Opinion-type questions mainly include two-sided questions and choice-type questions. The former refers to questions with the characteristics of [A or not A] such as "Is it right" and "Is it possible?" " and other questions with [A or B] characteristics. [0004] Existing reading comprehension models are often extractive reading comprehension models, which process information on questions and contexts. After word embeddin...

AI Technical Summary

Problems solved by technology

In this way, the candidate answers are independent of the context, and it is easier to solve [A, B, undetermined] type classification problems, but because the answers to opinion-type questions are often dynamic and require reasoning, most existing solutions tend to Because of classification, the idea of ​​classification cannot dynamically integrate the knowledge information of candidate viewpoints and reason the answers dynamically. In this way, it is often difficult to obtain ideal answers to dynamic choice questions.

Images