Interactive Matching Network for Multi-Turn Response Selection

This repository contains the source code and datasets for the CIKM 2019 paper Interactive Matching Network for Multi-Turn Response Selection in Retrieval-Based Chatbots by Gu et al. <br>

Our proposed Interactive Matching Network (IMN) has achieved a new state-of-the-art performance on four large-scale datasets that are publicly available for research on multi-turn conversation.

Model overview

<img src="image/model.png">

Results

<img src="image/UbuntuV1_V2.png"> <img src="image/Douban_Ecommerce.png">

Dependencies

Python 2.7 <br> Tensorflow 1.4.0

Datasets

Your can download the processed datasets used in our paper here and unzip it to the folder of data. <br> Ubuntu_V1 <br> Ubuntu_V2 <br> Douban <br> Ecommerce

Train a new model

Take Ubuntu_V1 as an example.

cd scripts
bash ubuntu_train.sh

The training process is recorded in log_train_IMN_UbuntuV1.txt file.

Test a trained model

bash ubuntu_test.sh

The testing process is recorded in log_test_IMN_UbuntuV1.txt file. And your can get a ubuntu_test_out.txt file which records scores for each context-response pair. Run the following command and you can compute the metric of Recall.

python compute_recall.py

Cite

If you use the code and datasets, please cite the following paper: "Interactive Matching Network for Multi-Turn Response Selection in Retrieval-Based Chatbots" Jia-Chen Gu, Zhen-Hua Ling, Quan Liu. CIKM (2019)

@inproceedings{Gu:2019:IMN:3357384.3358140,
 author = {Gu, Jia-Chen and 
           Ling, Zhen-Hua and 
           Liu, Quan},
 title = {Interactive Matching Network for Multi-Turn Response Selection in Retrieval-Based Chatbots},
 booktitle = {Proceedings of the 28th ACM International Conference on Information and Knowledge Management},
 series = {CIKM '19},
 year = {2019},
 isbn = {978-1-4503-6976-3},
 location = {Beijing, China},
 pages = {2321--2324},
 url = {http://doi.acm.org/10.1145/3357384.3358140},
 doi = {10.1145/3357384.3358140},
 acmid = {3358140},
 publisher = {ACM},
}