Dataset of homophobic messages on social networks in Colombia

The messages were collected using the API that Twitter provides upon invitation. During an observation window of one month and five days, we made around 38 requests of 100 messages. From these requests, 440 tweets segmented between negative and positive messages were selected, from which we captured ID, tweet (body's message), and date fields. To discern whether a message promoted homophobia, we chose those that contained colloquial terms of the country used to refer to the LGBTQ community or its members. We also considered those that included a word related to the subject and used offensive language. We should note that the double meaning is an aspect that makes classification difficult, since a word may or may not be offensive depending on the context. Due to it, we used terms similar to the negative ones for non-homophobic messages, but that expressed an opinion or support towards the community in question.