A Context-Constrained Sentence Modeling for Deception Detection in Real Interrogation Official Implementation

This repository is the official implementation of A Context-Constrained Sentence Modeling for Deception Detection in Real Interrogation.

1. How to start?

Extract emobase, BERT and the CTD feature by the tracing the following script.

1.1 Emobase

You need to install opensmile to extract emobase feature

feature_extract/emobase

1.2 BERT

Please install the dependency of ''pytorch-pretrained-bert 0.6.2'' then use this script to extract the BERT

feature_extract/bert

1.3 CTD

feature_extract/ctd

2. Training code

Trace run_multitask.py and models.py to get more information.

3. Testing code

Trace test_only.py to get more information.
We also provide our best model in the paper, please see BEST_MODEL.
We can provide some of our feature for testing only, please mail me to get more information.