2026-05-04T18:33:51.544Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11782 2026-01-06T17:55:29Z 3

Optimizing XGBoost for Online Tweets Harassment Detection Using Word2Vec and Modified Metaheuristics, Chapter in LNNS Lecture Notes in Networks and Systems: ICDSA 2025: Data Science and Applications, Springer, volume 1721 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11782 https://link.springer.com/chapter/10.1007/978-3-032-10753-4_29 D. Bulaja B. Radomirovic M. Tomic V. Zeljkovic T. Zivkovic M. Zivkovic L. Babic N. Bacanin The digital space provided by the Internet offers people a common medium for effortless interaction, exchanging opinions and establishing connections around the world. This kind of connectivity supports meaningful interactions and dissemination of various ideas beyond the scope of geographical boundaries. Nevertheless, these features that encourage such constructive engagement also create opportunities for mistreatment and harassment, particularly because of the veil of anonymity offered by the online world. The detrimental consequences of virtual abuse can be severe, often reflecting the harmful consequences linked to face-to-face intimidation and mobbing, ultimately resulting in considerable deterioration of the victim’s psychological well-being. To tackle this concern, artificial intelligence (AI) emerges as a viable solution due to its capability to swiftly interpret and evaluate immense volumes of data. While AI is traditionally used with numerical datasets, natural language processing (NLP) methods are uniquely tailored for handling written content and textual data, which makes them well-suited for harassment detection inside digital conversations. This research investigates the advantages of merging optimization metaheuristics with the XGBoost classification model to improve the detection of offensive comments on Twitter. To bolster the efficiency of the proposed technique, an adapted implementation of the renowned crayfish optimization algorithm metaheuristic is devised. The best produced classifiers through this methodology exhibit encouraging performance levels, attaining an accuracy of approximately 78.2%, highlighting their prospect for reliable online harassment detection. bookPart Springer, Cham 343 357 10.1007/978-3-032-10753-4_29 LNNS Lecture Notes in Networks and Systems: ICDSA 2025: Conference Proceedings of International Conference on Data Science and Applications, volume 1721