Novel Comment Spam Filtering Method on Youtube: Sentiment Analysis and Personality Recognition

摘要

The deeply entrenched use of Online Social Networks (OSNs), where millions of users share unconsciously any kind of personal data, offers a very attractive channel to attackers. They provide the possibility of sending spam messages through different channels (wall posts, comments, private messages). In this paper we propose a novel spam filtering method focused on social media spam. It aims to demonstrate that using sentiment analysis and personality recognition techniques, in order to analyze the content of the texts, the improvement of spam filtering results is possible. We add these features to each OSN spam both independently and jointly, and then we compare Bayesian spam filters with and without the new features in terms of the number of false positive and accuracy. At the end, the results of the top ten filtering classifiers have been improved, reducing also the number of false positives (26.69% on average), reaching an 82.55% of accuracy.