Kantepe

Preprocessing framework for Twitter bot detection

---

• authors: Mücahit Kantepe, Murat Can Ganiz

• link: https://ieeexplore.ieee.org/abstract/document/8093483

• file structure:

├── Twibot-20
│   ├── preprocess.py
│   └── train.py  # model on cresci-2015 and Twibot-20 
└── Twibot-22
    ├── preprocess.py
    └── train.py  # model on Twibot-22 

• implement details: In the original paper, the Gradient Boosted tree classification method has the best effect, but in our experiment, the effect of this method is significantly worse. The SVM method with the second best effect in the original paper is relatively better in our experiment, but the experiment is time-consuming. Since the core of this paper is feature selection rather than classification methods, we finally use the random forest method.

How to reproduce:

1. convert the raw dataset into standard format by running

   python preprocess.py

   this command will create related features in corresponding directory.

2. train random forest model by running:

   python train.py

   the final result will be showed.

Result:

random seed: 0, 100, 200, 300, 400

dataset		acc	precison	recall	f1
Cresci-2015	mean	0.975	0.813	0.753	0.782
Cresci-2015	std	0.013	0.014	0.012	0.014
Cresci-2017	mean	0.982	0.830	0.761	0.794
Cresci-2017	std	0.015	0.009	0.011	0.013
Twibot-20	mean	0.803	0.634	0.610	0.622
Twibot-20	std	0.043	0.021	0.019	0.021
Twibot-22	mean	0.764	0.786	0.468	0.587
Twibot-22	std	0.024	0.018	0.013	0.016

baseline	acc on Twibot-22	f1 on Twibot-22	type	tags
Kantepe et al.	0.764	0.587	F T	random forest