Augmenting Neural Response Generation with Context-Aware Topical Attention
Sequence-to-Sequence (Seq2Seq) models have witnessed a notable success in generating natural conversational exchanges. Notwithstanding the syntactically well-formed responses generated by these neural network models, they are prone to be acontextual, short and generic. In this work, we introduce a Topical Hierarchical Recurrent Encoder Decoder (THRED), a novel, fully data-driven, multi-turn response generation system intended to produce contextual and topic-aware responses. Our model is built upon the basic Seq2Seq model by augmenting it with a hierarchical joint attention mechanism that incorporates topical concepts and previous
interactions into the response generation. To train our model, we provide a clean and high-quality conversational dataset mined from Reddit comments. We evaluate THRED on two novel automated metrics, dubbed
Semantic Similarity and Response Echo Index, as well as with human evaluation. Our experiments demonstrate that the proposed model is able to generate more diverse and contextually relevant responses compared to the strong baselines.
Citation
N. Dziri,
E. Kamalloo,
K. Mathewson,
O. Zaiane.
"Augmenting Neural Response Generation with Context-Aware Topical Attention". ACL Workshop on NLP for Conversational AI, pp 18-31, July 2019.
| Keywords: |
|
| Category: |
In Workshop |
| Web Links: |
ACL |
| |
doi |
BibTeX
@misc{Dziri+al:19,
author = {Nouha Dziri and Ehsan Kamalloo and Kory Wallace Mathewson and Osmar
R. Zaiane},
title = {Augmenting Neural Response Generation with Context-Aware Topical
Attention},
Pages = {18-31},
booktitle = {ACL Workshop on NLP for Conversational AI},
year = 2019,
}Last Updated: September 15, 2020
Submitted by Sabina P 
