Method and system for optimizing machine reading comprehension ability based on hierarchical attention mechanism

Abstract

The invention discloses a method for optimizing a machine's reading comprehension ability based on a hierarchical attention mechanism, which uses multi-granularity input features to extract text information, and then uses an attention mechanism to interact with the extracted text information to obtain context information. Feature representation, and finally use new feature representations to get answers based on related questions. In the feature input stage, not only the regular vector representation of a simple word is input, but also the word vector of the pre-trained model of the word is input. At the same time, using the method of knowledge graph, the word vector of the external knowledge base named entity knowledge base NELL is also input. In the information interaction stage, various attention mechanisms are used. The question-aware text word representation is obtained by using the gated paragraph-to-question attention mechanism, and then through the linear self-matching mechanism, the question-aware text word can further perceive the text information before and after each text word, and then generate an answer.

Description

technical field [0001] The invention belongs to the technical field of natural language processing, and more specifically relates to a method and system for optimizing machine reading comprehension capabilities based on a hierarchical attention mechanism. Background technique [0002] With the emergence of large-scale high-quality data sets and the continuous improvement of modern computer computing power, machine reading comprehension technology has developed rapidly, and it has been verified that the accuracy rate is higher than that of humans in many fields. Today, machine reading comprehension technology is close to maturity and has been widely used in engineering practice, such as Taobao's 24-hour machine customer service, China Mobile's smart phone customer service, etc. Machine reading comprehension technology can replace manual 24-hour work and answer some common questions. The problem. [0003] Existing machine reading comprehension models usually use single word v...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a method and system for optimizing machine reading comprehension capabilities based on a hierarchical attention mechanism. Insufficient text information, semantic information, and lexical information provided, as well as the technical problem of low answer prediction accuracy due to the inability to fully extract the interaction information between paragraphs and text

Images