Humans are Main Culprit: Spotting Collusive Users Involved in Blackmarket Following Services

The repository contains the code and partial released dumps of the paper.

All the required functions are present in the repository.

Requirements

• Python 2.7.x To install the dependencies used in the code, you can use the requirements.txt file as follows -

pip install -r requirements.txt

Running the code

First cd SpotCU and then run:

bash prepare_and_test.sh

Motivation

Fake followers and collusive users, and their footprints: (a) Fake followers only follow blackmarket customers (synchronous); (b) Collusive users follow both blackmarket customers and genuine users (asynchronous); (c) Fake followers cause few noticeable spikes at outdegree distribution, whereas collusive and genuine users show multiple small spikes.

t-SNE

User projection in the two-dimensional embedding space. We use t-SNE plot to visualize the space. We randomly sample 200 collusive users and 200 genuine users for visualization.