Deception Indicators Prediction for Lie Detection in Dialogues

Existing techniques for identifying lies from the dialogue give a final decision without explanations. We design 4 deception indicators based on existing theory to lie detection. We design a new module, the deception indicators prediction module, to help lie detection in dialogues, and construct a l...

Full description

Saved in:
Bibliographic Details
Main Author: Ji, Chengwei
Format: Conference Proceeding
Language:English
Subjects:
BERT-LSTM
Cybernetics
Data models
Deception indicators
Feature extraction
Financial services
Human-machine systems
Lie detection
Oral communication
Predictive models
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Existing techniques for identifying lies from the dialogue give a final decision without explanations. We design 4 deception indicators based on existing theory to lie detection. We design a new module, the deception indicators prediction module, to help lie detection in dialogues, and construct a large labeled dataset from real financial service platforms. Based on this, we further propose a BERT-LSTM model which can both capture global information and preserve temporal features, making it highly advantageous for handling data with temporal characteristics for both deception indicator prediction and lie detection, such as conversations. Experimental results show the effectiveness of our model.
ISSN:2157-8982
DOI:10.1109/IHMSC62065.2024.00022