gongwu/nlpcc2018-task5-multiturnresponseselection

MultiTurnResponseSelection

Multi-Turn Response Selection

This repository contains code for the paper "Memory-Based Matching Models for Multi-turn Response Selection in Retrieval-Based Chatbots".

About this code

This is the code by ECNU submitted to nlpcc2018 Task5 sub1. The MBMN(MVM)+SMN+NLP model achieves 62.61% Precision score on the test set and ranks 1st among all the participants.

Installation

# download the repo
git clone https://github.com/gongwu/nlpcc2018-Task5-MultiTurnResponseSelection.git
# download the dataset
# run the model
python main/run_dialogue_SCNRMA.py

Results on Dev

	Model	Precision (%)
	NLP features	39.67
	SMN [ACL2017]	61.76
Single model	MBMN(MVM)	60.03
	MBMN(SMVM)	61.97
	MBMN(MVM)+SMN	62.11
Combined model	MBMN(SMVM)+SMN	62.08
	MBMN(MVM)+SMN+NLP	62.26
	MBMN(SMVM)+SMN+NLP	62.16