https://github.com/lukasdrews97/twitter_bot_detection
Twitter bot detection using graph neural networks.
https://github.com/lukasdrews97/twitter_bot_detection
bot-detection graph-neural-networks pytorch-geometric social-network twitter-bot-detection
Last synced: 19 days ago
JSON representation
Twitter bot detection using graph neural networks.
- Host: GitHub
- URL: https://github.com/lukasdrews97/twitter_bot_detection
- Owner: LukasDrews97
- Created: 2023-01-08T19:07:38.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2024-12-18T18:47:31.000Z (6 months ago)
- Last Synced: 2025-02-17T15:52:27.945Z (4 months ago)
- Topics: bot-detection, graph-neural-networks, pytorch-geometric, social-network, twitter-bot-detection
- Language: Python
- Homepage:
- Size: 148 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Twitter Bot Detection using Graph Neural Networks
Implementation of the BotRGCN architecture.
# Dataset Statistics
The model was trained on the TwiBot-22 dataset.
| **Unit** | **Human** | **Bot** | **Total** |
|------------------------|-------------|-----------|-----------|
| Users (all) | 860.057 | 139.943 | 1.000.000 |
| Users (min. 1 tweet) | 818.613 | 115.259 | 933.872 |
| Tweet | 81.250.102 | 6.967.355 | 88.217.457|
| Following relation | 1.038.302 | 78.353 | 1.116.655 |
| Follower relation | 2.383.574 | 243.405 | 2.626.979 |
| Retweet relation | 1.482.911 | 97.732 | 1.580.643 |
| Hashtags | 56.353.776 | 9.646.857 | 66.000.633|
# User Relations
A user relation is an interaction between two users. When training Graph Neural Networks, relations are modeled as edges between nodes.
The following relations exist in the TwiBot-22 dataset:
- Follower relation (F): User A follows User B
- Following relation (F): User A is followed by User B
The following relations were derived from the TwiBot-22 dataset:
- Retweet relation (R): User A retweeted a tweet from User B
- Co-Retweet relation (Co-R): User A and User B retweet the same tweet
- Co-Hashtag relation (Co-H): User A and User B use the same hashatg
# Training Results
The model was trained using different combinations of user relations and achieves state-of-the-art performance.
| **Configuration** | **Accuracy** | **F1-Score** |
|--------------------|--------------------|--------------------|
| F-F | 77.0 ± 0.18 | 50.4 ± 0.74 |
| F-F, R | 79.0 ± 0.88 | 57.4 ± 0.31 |
| F-F, Co-R | 79.0 ± 0.17 | 57.1 ± 0.62 |
| F-F, R, Co-R | 79.0 ± 0.20 | 57.7 ± 0.25 |
| F-F, R, Co-R, Co-H | 79.2 ± 0.10 | 57.9 ± 0.51 |