Social networks represent a significant threat to users who are exposed to many risks and potential attacks. One of such threats is aggressive comments, which can produce long-term harm to victims, in the more accurate cases they can lead to suicide. This track focuses on the detection of aggressive comments in Twitter, a topic that has not been widely studied in the community. Participants will have to develop methods to determine whether a tweet is aggressive or not. This challenging task is further complicated by the fact that tweets come from Mexican users and from with a variety of backgrounds, making it a quite challenging (yet realistic and with high impact) problem.
The data set for this track was collected between August and November 2017 according to the following methodology. Firstly, tweets were collected based on a fixed vocabulary extracted from a dictionary of “Mexicanisms.” Mainly we considered the subset of words classified as “vulgar” or “insult” and searched for tweets containing at least one of these words. Then, tweets were manually labeled by two persons as aggressive or non-aggressive. Taggers were provided with a labeling manual based on the premise that an offensive message is characterized by disparaging or humiliating a person or a group of persons. Therefore, an offensive message may contain some of the following elements: nicknames (assigned to the person/persons the message is addressed, alluding to a disability or defect), jokes (as long as they intend to humiliate or attack), derogatory adjectives (used with the intention of humiliating) and profanities (bad words or high-sounding expressions used to attack a person).