A machine learning-based malicious bot detection framework for trend-centric twitter stream

Abstract

There is an increase in evidence about the generation of social media content by autonomous entities called bots. These automated entities may be computerized algorithms or a human-assisted account to boost a trend or control political campaigns. People do not create bots due to malicious intent always. Several bots are benign and often play helpful roles. Detection of these bots have many implications. Most techniques available to date perform bot detection at the account level. This paper aims to uncover social bots that are active in trending events on Twitter. We have designed an automated malicious bot detection framework to serve this purpose. Our main contributions include converting unlabelled Twitter extracted trend-centric dataset to labelled dataset using semi-supervised machine learning techniques. Selection of attributes useful for bot classification using Entropy and Information Gain. Formulation of rule hierarchy after feature selection and threshold estimation of respective features. We have finalized the suspected bots after applying the rule hierarchically on the initial dataset. In addition to this, validation of the alleged bots given by the proposed framework is performed by plotting timeline tweets of suspected users, acts as an enhancement. This model is useful for bot detection in trend-centric Twitter stream. The research carried out will serve as an advancement to the successful detection of bots in trend-centric datasets.

Subject Classification:

• 91D30
• 97M70
• 68Q87

Keywords:

• Social media
• Social media bots
• Machine learning classifiers
• Trend-centric
• Malicious bots
• Bot detection techniques