Application-Oriented Approach for Detecting Cyberaggression in Social Media

摘要

The paper discuses and demonstrates the use of named-entity recognition for automatic hate speech detection. Our approach also addresses the design of models to map storylines and social anchors. They provide valuable background information for the analysis and correct classification of the brief statements used in social media. Furthermore, named-entity recognition can help to tackle the specifics of the language style often used in hate tweets, a style that differs from regular language in deliberate and unintentional misspellings, strange abbreviations and interpunctuations, and the use of symbols. We implemented a prototype for our approach that automatically analyzes tweets along storylines. It operates on a series of bags of words containing names of persons, locations, characteristic words for insults, threats, and phenomena reflected in social anchors. We demonstrate our approach using a collection of German tweets that address the vitally discussed topic "refugees" in Germany.