DAWN

KO | EN

gitlite — search

#javascript #hacktoberfest #ai #typescript #llm #go #golang #machine-learning #rust #deep-learning #docker #nodejs #awesome #openai

(★ 20)

The official source code of the paper "A New DAWN for Fake News Detection: Exploiting Engagement Earliness for Temporality-aware Evaluation" (TIST 2026)

DAWN 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# [TIST 26] A New DAWN for Fake News Detection: Exploiting Engagement Earliness for Temporality-aware Evaluation

This repository contains the official implementation of **DAWN**, a method for **D**etecting fake news via e**A**rliness-guided re**W**eighti**N**g.

This work was originally published at WSDM 2025 as "[Revisiting Fake News Detection: Towards Temporality-aware Evaluation by Leveraging Engagement Earliness](https://dl.acm.org/doi/pdf/10.1145/3701551.3703524)", and has been subsequently extended into the journal version:

[**"A New DAWN for Fake News Detection: Exploiting Engagement Earliness for Temporality-aware Evaluation"**](https://dl.acm.org/doi/pdf/10.1145/3815193), accepted to ACM Transactions on Intelligent Systems and Technology (TIST), 2026.

Junghoon Kim*, Junmo Lee*, Yeonjun In, Kanghoon Yoon, and Chanyoung Park

\* Both authors contributed equally to this research.

## Abstract
Social graph-based fake news detection aims to identify news articles containing false information by utilizing social contexts, e.g., user information, tweets and comments. However, conventional methods are evaluated under less realistic scenarios, where the model has access to future knowledge on article-related and context-related data during training. In this work, we newly formalize a more realistic evaluation scheme that mimics real-world scenarios, where the data is ***temporality-aware*** and the detection model can only be trained on data collected up to a certain point in time. We show that the discriminative capabilities of conventional methods decrease sharply under this new setting, and further propose **DAWN**, a method more applicable to such scenarios. Our empirical findings indicate that later engagements (e.g., consuming or reposting news) contribute more to noisy edges that link real news-fake news pairs in the social graph. Motivated by this, we utilize feature representations of engagement earliness to guide an edge weight estimator to suppress the weights of such noisy edges, thereby enhancing the detection performance of **DAWN**. Through extensive experiments, we demonstrate that **DAWN** outperforms existing fake news detection methods under real-world environments.

## Overall Architecture
<p align="center"><img width="600" src="./images/architecture.png"></p>

## Requirements
- python=3.9.18
- pytorch=1.12.1
- scikit-learn=1.3.0
- numpy=1.26.3
- scipy=1.11.4 
- pyg=2.4.0 (torch-geometric)
- twitter=1.19.6

## Data Download/Customization

Our work is based on the `PolitiFact` and `GossipCop` datasets from the [FakeNewsNet benchmark](https://github.com/KaiDMML/FakeNewsNet).

You can download our paper's implementation version or directly customize the dataset with different hyperparameters. Here's the following procedures.

### 1. Use paper's implementation version

- You can download paper's pre-determined version of two datasets in this drive [link](https://drive.google.com/drive/folders/1YbHgef66Jkf_0EaT7GTWm8VmEb3eSEi7?usp=sharing).
- Download each folder and place it in the current directory. (We upload our datasets into drive link due to large file size.)

### 2. Customize the dataset with different hyperparameters

- You can generate a customized dataset with different hyperparameter settings (e.g. deadline, threshold) with ```construct_data.py``` files.

- First, `obj_base_data.pkl` files should be located in the `main/obj/` directory.
  * Please download the files using the [link](https://drive.google.com/drive/folders/1YbHgef66Jkf_0EaT7GTWm8VmEb3eSEi7?usp=sharing), and use the `obj_base_data.pkl` file from the downloaded files.
  * For example, relative path for gossipcop `pkl` file should follow the form as `main/gossipcop/gossipcop_base_data.pkl`

- Second, select your desired hyperparameters as follows.

* #### Short Explanations for dataset hyperparameters

* `--obj:` name of dataset (e.g. `'politifact'`, `'gossipcop'`)

* `--train_ratio:` training ratio for experiments (e.g. `0.7`)

* `--valid_ratio:` valid ratio for experiments (e.g. `0.2`)

* `--u_thres:` threshold for active user (e.g. `3`)

```bash
python construct_data.py --obj 'gossipcop' --train_ratio 0.7 --valid_ratio 0.2 --u_thres 3
```

- Third, check the directory to see if the full data `pkl` file is well made.
  * Example of relative path for gossipcop full data file : `main/gossipcop/gossipcop_t3_train_ratio0.7_valid_ratio0.1_full.pkl`

## How to Run?
You can run the model our paper's best performance hyperparameter set or your desired hyperparameter set.

### 1. Use paper's implementation version

- You can use shell file for running ```run.py``` with our best performance hyperparameter set.
```bash
sh run.sh
```

### 2. Use customized version

- You can use your customized dataset and select your desired hyperparameters as follows.

* #### Short Explanations for feature construction/model hyperparameters

* `--dead_criterion:` deadline for engagements ($t_d$ in our paper) / should follow the form as `'12hrs'` or `'30m'`

* `--early_thres:` threshold for users ($thres_u$ in our paper)

* `--hidden_dim:` size of gnn hidden layer dimension

* `--epochs:`  number of epochs for training DAWN

* `--lr:` learning rate for training the model

* `--margin:` margin for ranking loss

* `--sampling_num:` number of sampling ( $K$ in our paper )

* `--alpha:` balancing the contribution of the ranking loss ( $alpha$ in our paper )

```bash
python run.py --dataset_name 'gossipcop' --gpu 5 --epochs 1000 --iters 5 --dead_criterion "12hrs" --early_thres 0.7 --sampling_num 10000 --margin 0.0 --alpha 0.3
```

### Models
`model.py`: implementation file of DAWN

### Others
`TweetTimeline.txt`: txt file for calculating real-world time from tweet id

`TimestampEstimator.py`: python file for converting tweet id to real-world time ( adapted from TweetedAt )

`utils.py`: utils for DAWN

### Cite
- If you find ``DAWN`` useful in your research, please cite the following paper:
  - Junghoon Kim*, Junmo Lee*, Yeonjin In, Kanghoon Yoon, and Chanyoung Park. "A New DAWN for Fake News Detection: Exploiting Engagement Earliness for Temporality-aware Evaluation" TIST 2026.
  - Bibtex

```
@article{kim2026new,
  title={A New DAWN for Fake News Detection: Exploiting Engagement Earliness for Temporality-aware Evaluation},
  author={Kim, Junghoon and Lee, Junmo and In, Yeonjun and Yoon, Kanghoon and Park, Chanyoung},
  journal={ACM Transactions on Intelligent Systems and Technology},
  year={2026},
  publisher={ACM New York, NY}
}
```
- The earlier conference version appeared at WSDM:

```
@inproceedings{kim2025revisiting,
  title={Revisiting fake news detection: Towards temporality-aware evaluation by leveraging engagement earliness},
  author={Kim, Junghoon and Lee, Junmo and In, Yeonjun and Yoon, Kanghoon and Park, Chanyoung},
  booktitle={Proceedings of the Eighteenth ACM International Conference on Web Search and Data Mining},
  pages={559--567},
  year={2025}
}
```

DAWN

DAWN 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

DAWN

DAWN 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README