Deep learning question and answer reasoning method and device based on graph attention mechanism

Abstract

The invention discloses a deep learning question and answer reasoning method and device based on a graph attention mechanism. The invention provides a reasoning model AGTF based on a graph point ride attention algorithm, a mixed model fusing ALBERT and a graph attention mechanism (GAT) is provided for the multi-hop problem in questions and answers, the model comprises a coding and decoding layer and a graph neural network prediction layer, and an experiment result shows that compared with an existing multi-hop question and answer reasoning algorithm, the algorithm has the advantages that the algorithm is simple, the efficiency is high, and the efficiency is high. The AGTF model effectively improves the reasoning ability of multi-hop questions and answers.

Description

technical field [0001] The invention belongs to the technical field of computer applications, and relates to a question and answer reasoning method integrating ALBERT and graph attention mechanism. Background technique [0002] The ability to reason and reason about natural language is an important aspect of artificial intelligence. The automatic question answering task provides a quantifiable and objective method to test the reasoning ability of artificial intelligence systems, and is gradually becoming a new trend of natural interaction between humans and machines, which can more accurately understand user questions described in natural language, and It will become a new form of the next-generation search engine by returning more accurate answers to users based on their true intentions. [0003] Question answering has always been a hot topic in the field of natural language processing. QA provides a quantitative method for evaluating the ability of NLP systems in language...

AI Technical Summary

Problems solved by technology

[0003] Question answering has always been a hot topic in the field of natural language processing. QA provides a quantitative method for evaluating the ability of NLP systems in language understanding and reasoning. The development of deep learning models has made great progress in the field of machine reading comprehension and question answering. , and even surpassed humans on single-paragraph question answering benchmarks including SQuad, but most of the previous work focused on finding evidence and answers from a single paragraph, and rarely tested the deep reasoning ability of the underlying model. The question-answering gap between humans faces the challenge of improving the reasoning ability of the model. The single-segment question-answering model tends to find the answer in the sentence that matches the question, does not involve complex reasoning and is still lacking when a single document is not enough to find the correct answer. Ability to reason about multiple documents

Images