Dehghan

Detecting Bots in Social-Networks Using Node and Structural Embeddings

---

• authors: Ashkan Dehghan, Kinga Siuta, Agata Skorupka, Akshat Dubey, Andrei Betlen, David Miller,Wei Xu, Bogumił Kaminski,Paweł Prałat

• link: https://www.researchsquare.com/article/rs-1428343/latest.pdf

• file structure:

├── network.py # generate graph features
├── profile_features.py # generate category features
├── nlp_features.py # generate text features
├── all_fea.py # check to make sure all features needed has been generated
├── **__fea*.py # features using specific model
└── main_xgboost.py # train model on every dataset

• implement details:
  • We did not reimplement the rest of the algorithms on cresci-2015 due to the limiation of the computational resources and the lack of the efficiency of the aforementioned algorithms on such a large dataset with millions of edges.

---

How to reproduce:

The data has been preprocessed and stored in folders e.g./cresci-2015

first run all_fea.py to generate the total features used for training remember to change the file path according to the dataset name

dataset='Twibot-20'

---

then you can use the feature generated ,change the dataset name ,and run main_xgboost.py Check the results in results/dataset.log

Result:

dataset		acc	precison	recall	f1
Twibot-22	mean	-	-	-	-
Twibot-20_all	mean	0.8604	0.9472	0.8219	0.8801
Twibot-20_Deepwalk	mean	0.8634	0.9400	0.8300	0.8816
Twibot-20_Node2vec	mean	0.8607	0.9425	0.8798	0.8718
Twibot-20_Role2Vec	mean	0.8607	0.9484	0.8261	0.8805
Twibot-20_RolX	mean	0.8653	0.9313	0.8378	0.8820
Twibot-20_Struc2Vec	mean	0.8617	0.9366	0.8298	0.8799
Twibot-20_GraphWave	mean	0.8668	0.9331	0.6311	0.7620
cresci-2015_GraphWave	mean	0.6206	0.9615	0.8388	0.8834
cresci-2015_Node2Vec	mean	0.6318	0.9615	0.8388	0.7743