Addressee and Response Selection System

This repo contains Theano implementations of the models described in the following papers:

Addressee and Response Selection for Multi-Party Conversation. EMNLP 2016

Addressee and Response Selection Task

Dataset

The dataset can be downloaded at data/input.

• When you use /data/input/train-data.cand-10.1/2.gz, please concatenate these two files into one: zcat train-data.cand-10.1.gz train-data.cand-10.2.gz | gzip > train-data-cand-10.gz

Dependencies

To run the code, you need the following extra packages installed:

• Numpy and Theano

Usage

1. Clone this repo
2. Move to the directory code/: cd code/
3. Run python -m adr_res_selection.main.main --help to see all running options

Example Comand

• Static Model: python -m adr_res_selection.main.main -mode train --train_data ../data/input/train-data.cand-2.gz --dev_data ../data/input/dev-data.cand-2.gz --test_data ../data/input/test-data.cand-2.gz --model static
• Dynamic Model: python -m adr_res_selection.main.main -mode train --train_data ../data/input/train-data.cand-2.gz --dev_data ../data/input/dev-data.cand-2.gz --test_data ../data/input/test-data.cand-2.gz --model dynamic