Sentiment and time-series analysis of direct-message conversations

Abstract

Social media and mobile communications in general are an extremely rich source of digital forensic information. We present our new framework for analysing this resource with an innovative combination of time series and text mining methods. The framework is intended to create a tool to analyse and operationally summarise extended trails of social media messages, thus enabling investigators for the first time to drill down into specific moments at which sentiment analysis has detected a change of tone indicative of a particularly strong and significant response. Crucially, the method will give investigators an opportunity to reduce the time and resource commitment required for ongoing and hands-on analysis of digital communications on media such as Texts/SMS, WhatsApp and Messenger.