LGCM

This is the official repository for the Findings of EACL 2024 paper "Local and Global Contexts for Conversation".

Environment

• python==3.6.8
• torch==1.4.0
• transformers==3.0.2

Usage

Data preparation

Tokenizer

We use GPT2 vocabulary in our experiments. To prepare vocabulary files, please:

• download gpt2-vocab.json from here, rename it to vocab.json, and move it to the folder ./gpt2_vocab/
• download gpt2-merges.txt from here, rename it to merges.txt, and move it to the folder ./gpt2_vocab/

Datasets

We have trained LGCM on three public available dalogue datasets:

• PersonaChat
• DalyDialog
• MultiWOZ

After downloading raw data, please run scripts in ./prepare_data/ to preprocess data.

Training

• PersonaChat: bash scripts/train_personachat.sh
• DailyDialog: bash scripts/train_dailydialog.sh
• MultiWOZ: bash scripts/train_multiwoz.sh

Evaluation

• PersonaChat: bash scripts/evaluate_personachat.sh
• DailyDialog: bash scripts/evaluate_dailydialog.sh
• MultiWOZ: bash scripts/evaluate_multiwoz.sh