Abstract:
Proper response selection is an important challenge for a meaningful multi-turn dialogue. To this end, not only the coherence among the whole dialogue but also the intera...Show MoreMetadata
Abstract:
Proper response selection is an important challenge for a meaningful multi-turn dialogue. To this end, not only the coherence among the whole dialogue but also the interaction between utterance in adjacent turns need to be properly employed as the context for response selection. In this paper, we propose a deep hybrid network (DHN) to distill such contextual information. First, we match the response with each utterance and filter internal noises with recurrent neural networks. Second, several deep convolutional blocks perform as a feature extractor and output a matching vector to be fused into a final matching score. During this period, complex contextual information across the whole conversation can be thoroughly blended and captured. The empirical study on two commonly used public datasets has shown the proposed model's potential.
Published in: ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
Date of Conference: 12-17 May 2019
Date Added to IEEE Xplore: 17 April 2019
ISBN Information: