Week-Signal Analysis Based on Expert Estimations and Big Data Tools

The aim of the research is an analysis of unstructured textual data (mainly from social media and mass-media) to identify and detect weak signals that could predict future events in context of protest activity and riots, trying to indicate them beforehand. In this work we have tried to design and implement a weak signal detection approach on the base of feature analysis on a variety of datasets from social media by applying a machine learning methods. We used a sample use case with a dataset which was manually labeled by expert. Semantic and sentiment features to the traditional syntactic features were combined. Through the procedure, we also tried to give a detailed analysis on the performance of our approach, and showed it for particular dataset and how it could work in real-time analytics. According to the task defined above it is necessary to solve the followng problems tasks: theoretical study of the topic; preparation of datasets for supervised learning for weak signal detection and identification; building predictive models for future events on the basis of weak signal prediction; develop utilities for weak signal detection of riots and protests as a kind of monitoring system.