Overview / Usage

Problem: Twitter bots are becoming harder to identify, as they’re finding new ways to blend in with ‘real’ user accounts. Previous twitter bot techniques are not able to identify bots, as they previously did.
Solution: Implement a machine learning classifier capable of distinguishing between twitter bots (specifically, social spambot) and real, verified human accounts based on (1) user’s account information and (2) their tweets.

Methodology / Approach

Users - Tested with 8386 examples (3474 ‘Genuine’ class & 4912 ‘Spambot’ class).
Tweets - Tested with 1021 examples (743 ‘Genuine’ class & 278 ‘Spambot’ class).
Implemented Random Forests classifier and 10 Fold Cross-Validation.
Implemented classifier and validation method in Python, and compared with results from Weka.

Technologies Used

Python, Twitter API, Weka, Scikit-learn, Pandas