Detection of Bot Accounts in a Twitter Corpus: Author Profiling of Social Media Users as Human vs. Nonhuman

This paper presents a successful series of experiments on the detection of SPAMBOTS in Twitter, based on the use of linguistic features. For these experiments, we built a small corpus and classified its contents with the help of human annotators, who achieved a high rate of agreement. We identified linguistic features previously tested in the literature and adapted them to the language and contents of our database. High accuracy, (90%), was achieved in the spambot detection task. Our best results were obtained with a very small feature set produced with automatic reduction techniques. This outcome supports our contention that feature reduction is crucial in text classification tasks. All experiments were conducted by means of software packages with GUIs that do not require programming skills. Our results highlight the fact that language experts can, with a little training, utilize their knowledge and expertise in the very important fight against malicious technologies.