Skip to content
/ REAL Public

This repo is the official implementation of "REAL: Retrieval-Augmented Prototype Alignment for Improved Fake News Video Detection“, accepted by ICME 2025.

License

Notifications You must be signed in to change notification settings

Jian-Lang/REAL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

REAL: Retrieval-Augmented Prototype Alignment Framework

This repo provides a official implementation for paper: REAL: Retrieval-Augmented Prototype Alignment for Improved Fake News Video Detection, accepted by ICME 2025.

Abstract

Detecting fake news videos has emerged as a critical task due to their profound implications in politics, finance, and public health. However, existing methods often fail to distinguish real videos from their subtly manipulated counterparts, resulting in suboptimal performance. To address this limitation, we propose REAL, a novel model-agnostic REtrieval-Augmented prototype-aLignment framework. REAL first introduces an LLM-driven video retriever to identify contextually relevant samples for a given target video. Subsequently, a dual-prototype aligner is carefully developed to model two distinct prototypes: one representing authentic patterns from retrieved real news videos and the other encapsulating manipulation-specific patterns from fake samples. By aligning the target video’s representations with its ground-truth prototype while distancing them from the opposing prototype, the aligner captures manipulation-aware representations capable of detecting even subtle video manipulations. Finally, these enriched representations are seamlessly integrated into existing detection models in a plug-and-play manner. Extensive experiments on three benchmarks demonstrate that REAL largely enhances the detection ability of existing methods.

Framework

image

Source Code Structure

├── data    # dataset path
│   ├── FakeSV
│   ├── FakeTT
│   └── FVC
├── preprocess  # code for prepocessing data
│   ├── make_retrieval_tensor.py
│   ├── generate_caption_BLIP.py
│   ├── generate_query_text.py
├── retrieve    # code of conducting retrieval
│   └──conduct_retrieval.py
├── run         # script for preprocess and retrival
├── src         # code of model arch and training
│   ├── main.py     # main code for training 
│   ├── model
│   │   ├──Base
│   │   └──SVFEND    # implementation of SVFEND w/ REAL
└── └── utils

Dataset

We provide video IDs for each dataset splits. Due to copyright restrictions, the raw datasets are not included. You can obtain the datasets from their respective original project sites.

Usage

Requirement

To set up the environment, run the following commands:

conda create --name REAL python=3.12
conda activate REAL
pip install -r requirements.txt

Preprocess

  1. Download datasets and store them in data presented in Source Code Structure, and save videos to videos in corresponding datasetpath.
  2. For video dataset, save data.jsonl in each dataset path, with each line including vid, title, ocr, transcript, and label.
  3. Run following codes to prepocess data:
bash run/retrieve.sh  # preprocess data and conduct retrieval
bash run/preprocess.sh  # preprocess data for SVFEND w/ REAL

Run

python src/main.py --config-name SVFEND_FakeSV.yaml     # run SVFEND w/ REAL on FakeSV
python src/main.py --config-name SVFEND_FakeTT.yaml     # run SVFEND w/ REAL on FakeTT
python src/main.py --config-name SVFEND_FVC.yaml        # run SVFEND w/ REAL on FVC

Citation

If you find our research useful, please cite this paper:

@inproceedings{li2025real,
	author = {Li, Yili and Lang, Jian and Hong, Rongpei and Chen, Qing and Cheng, Zhangtao and Chen, Jia and Zhong, Ting and Zhou, Fan},
	booktitle = {IEEE International Conference on Multimedia and Expo (ICME)},
	year = {2025},
	organization = {IEEE},
	title = {REAL: Retrieval-Augmented Prototype Alignment for Improved Fake News Video Detection},
}

About

This repo is the official implementation of "REAL: Retrieval-Augmented Prototype Alignment for Improved Fake News Video Detection“, accepted by ICME 2025.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published