Dialog System Technology Challenges 7 (DSTC7) Track 1

Implementation of the model for Track 1 of 7th Edition of the Dialog System Technology Challenges with Tensorflow

This repository contains an implementation with Tensorflow of the model presented in the paper Building Sequential Inference Models for End-to-End Response Selection by Gu et al. in AAAI 2019 Workshop on DSTC7.
In this challenge, the model has achieved the results of rank 2nd on Ubuntu dataset and rank 3rd on Flex dataset.

Dependencies

Python 2.7
Tensorflow 1.4.0

Data Preparation

Step 1. Refer to the official website DSTC7 to download the dataset.

Step 2. Create a vocabulary file composed of words in the datast and name it as "vocab.txt".

Step 3. Download the Glove embeddings and train word embeddings with Word2Vec. Then combine these two embedding files as proposed in the paper by running the following commands.

cd data
python train_word2vec.py
python combine_emb.py

Running the scripts

Train a new model

cd scripts
bash train.sh

The training process and results are in log.txt file.

Test a trained model

bash test.sh

The test results are in log_test.txt file.