A System to Filter out Unwanted Social Media Content in Real-time on iPhones

Anonymous

Abstract: Social media users are often harassed. This paper presents a patented system to filter out harassing content before it reaches the recipient. Our first version is for the iPhone. To detect harassment, we adopted sentiment analysis with a supervised learning approach that combines Machine Learning (ML) text classifiers with a lexicon approach that provides a feedback loop to retrain the ML model with unknown terms. Because good data is essential to obtain the best output of any system, we focused on validating our labeled data. Our results on static and real-time data have an accuracy of, respectively, 90% and 94%. Our labeled data validation allows us to correct labels; we also realized the need to increase the number of sets in our lexicons. Our prototype demonstrates that we are able to build an AI infrastructure to filter out harassment on an iPhone in real-time with good results.