Open AccessMulti-Granularity Matching Network for Multi-Paragraph Machine Reading Comprehension
Author(s) -
Xu Chen
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1871/1/012091
Subject(s) - paragraph , granularity , computer science , matching (statistics) , reading comprehension , reading (process) , artificial intelligence , comprehension , natural language processing , word (group theory) , encoding (memory) , key (lock) , linguistics , world wide web , mathematics , programming language , statistics , philosophy , computer security
Multi-paragraph aims to allow machines to read multiple paragraphs and infer the answer to the given question. Usually, the model needs to use the selector to narrow the range of candidate paragraphs, and then use the reader to find the answer. For the selector, the previous work generally matches the question and the paragraph on the word level and the paragraph level. In this paper, we study the encoding and matching algorithm at the chunk level, which is supposed to make our model more accurate. Combining it with gated self matching mechanism, we design a multi-granularity matching network as the selector. The experimental results show that our model achieves competitive performance on Quasar-T.