TT-Net: Topic Transfer-Based Neural Network for Conversational Reading Comprehension

Conversational machine reading comprehension (MRC) is a new question answering task, which is more challenging compared to traditional single-turn MRC since it requires a better understanding of conversation history. In this paper, a novel neural network model for conversational reading comprehensio...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2019, Vol.7, p.116696-116705
Main Authors: Yingjie, Gu, Xiaolin, Gui, Defu, Li
Format: Article
Language:English
Subjects:
Conversational reading comprehension
Feature extraction
neural network
Neural networks
Questions
Reading comprehension
temporal convolutional network
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Conversational machine reading comprehension (MRC) is a new question answering task, which is more challenging compared to traditional single-turn MRC since it requires a better understanding of conversation history. In this paper, a novel neural network model for conversational reading comprehension, namely TT-Net, is proposed, which is capable of capturing topic transfer features using temporal convolutional network (TCN) in the dialog. The TT-Block packaged by the BiLSTM, TCN and Self-attention mechanism is presented to extract topic transfer features between questions. Our model is evaluated on the CoQA benchmark dataset compared with several baseline models including the strong baseline model named FlowQA. The results show that the model outperforms the baseline models: BiDAF++ by 7.6% and FlowQA by 0.7%, especially in children's story domain our model promotes FlowQA's performance by 3.9%, which indicates that the TT-Net contributes to a decent promotion for conversational reading comprehension.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2019.2936011