# Sequential Match Network: A New Architecture for Multi-turn Response Selection in Retrieval-based Chatbots

**Authors:** Yu Wu, Wei Wu, Ming Zhou, Zhoujun Li  
**arXiv:** [1612.01627](https://arxiv.org/abs/1612.01627)  
**Date:** 2016-12-06  
**Summary page:** https://www.summarizepaper.com/en/paper/1612.01627v1/

## Key points

- Authors propose a new session-based matching model for response selection in multi-turn conversation in retrieval-based chatbots
- Model addresses limitations of existing works by considering relationships among utterances and important contextual information
- Model matches response with each utterance on multiple granularities and distills important matching information using convolution and pooling operations
- Matching vectors are accumulated in chronological order through a recurrent neural network (RNN) to model relationships among utterances
- Final matching score is calculated using hidden states of the RNN
- Empirical studies show that the proposed model outperforms state-of-the-art methods for response selection in multi-turn conversation
- Research contributes to enhancing performance of retrieval-based chatbots by considering relationships among utterances and important contextual information during response selection

## Summary

The authors of this paper propose a new session-based matching model for response selection in multi-turn conversation in retrieval-based chatbots. They address the limitations of existing works that either ignore relationships among utterances or miss important information in context when matching a response with a highly abstract context vector. The proposed model first matches a response with each utterance on multiple granularities and distills important matching information from each pair using convolution and pooling operations. These matching vectors are then accumulated in chronological order through a recurrent neural network (RNN) that models the relationships among the utterances. The final matching score is calculated using the hidden states of the RNN. Empirical studies conducted on two public datasets demonstrate that the proposed model outperforms state-of-the-art methods for response selection in multi-turn conversation, indicating its effectiveness in capturing contextual information and leveraging relationships among utterances to improve response selection accuracy. This research contributes to enhancing the performance of retrieval-based chatbots by considering both the relationships among utterances and important contextual information during response selection.