Spammers on social network detection
January 18, 2019 · View on GitHub
Feature extraction, classification , and prediction of spammers on social network
In this project we focus on detecting spammers in a social network. The ultimate purpose of this project to identify (i.e., classify) the spammer users based on their relational and non-relational features.
See the report for more details.Running the code
The dataset represent users from tagged.com social network. The full dataset is provided in : https://linqs-data.soe.ucsc.edu/public/social_spammer/.
Please think to change the name of the csv files to be able to run the code. These are the name of the different datasets we worked on :
- usersdata.csv : this file contain the different users and they content features.
- filtered_relations_1.csv : This file contains the edges of the relation 1 between the nodes.
- filtered_relations_2.csv : This file contains the edges of the relation 2 between the nodes.
- filtered_relations_4.csv : This file contains the edges of the relation 4 between the nodes.
- filtered_relations.csv : This file contains the edges of the relation 5 between the nodes.
- filtered_relations_7.csv : This file contains the edges of the relation 7 between the nodes.
These filtered relations files should be generated using the spammer_subnetwork.ipynb after downloading the full dataset from the link above.
The repository contains four jupyter notebooks :
- spammer_subnetwork.ipynb : (©) Eda Bayram : Please filter the original huge file based on the relation. Create relation 1 2 4 5 and 7: these are the relations we worked on for this project.
- feature_extraction.ipynb : This Notebook allows to compute the graph based features and create a final dataset containing the nodes of the network and the calculated features (content based and graph based).
- feature_exploration.upynb : This notebook try to better understand the computed features in details.
- Classification.ipynb : This notebook allows to tune hyperparameters for different supervised learning algorithm on the dataset provided by the first notebook in order to classify the nodes beetween spammers and non spammers.
More details
See the project report provided also in the repository. The slides presented is also provided.
Authors