TRACT: Tweets Reporting Abuse Classification Task Corpus

Out of many social media sites, Twitter has some unique characteristics which attract diverse categories of people. Twitter has been utilized as an important source of user-generated data that can provide unique insights into population. Many of these studies involve retrieving tweets for sentiment analysis, named entity recognition and disambiguation tasks. In recent decades we have noticed a considerable increase in reports or confession posts of abuse victims on twitter. Most of the time victims do not report it to their guardians or the concerned authorities. Part of these victims tweets about their incident to let go of pain and suffering or as a cry for help. Identifying such reports is challenging, to address such an important task we develop and release a first small-scaled corpus on Tweet Reportings (TRACT). To retrieve the small number of tweets that mention reporting of abuse we mined from millions of tweets in our database, The approach relies on a lexicon and lexical variants from a set of lexicon entries manually collected after extensive research. Tweets containing these lexicons and tweets containing lexicons similar to these lexicons are extracted