2026-08-21T21:16:14.738Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:12153 2026-08-19T21:53:44Z 3

An Optimized XGBoost Approach for Racism Detection in Twitter Comments Employing Gensim and Modified Variable Neighborhood Search, LNNS Lecture Notes in Networks and Systems: ICDPN 2025: Data Processing and Networking, Springer, volume 2025 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/12153 https://link.springer.com/chapter/10.1007/978-3-032-28616-1_22 J. Maricic D. Cvetkovic M. Grubjesic T. Zivkovic M. Zivkovic L. Anicin S. Anetic N. Bacanin The explosive expansion of social networks has reshaped the ways individuals interact and disseminate information, yet it has simultaneously intensified the circulation of racist commentary and hostile discourse. Elements such as concealed user identities, reduced inhibition in online environments, and insufficient oversight have magnified this issue, further sustaining persistent patterns of discriminatory expression. To confront this challenge, the present study proposes an extensive framework that integrates artificial intelligence, sophisticated natural language processing techniques, and machine learning classifiers to more precisely identify racist material. Employing Gensim-based preprocessing to minimize noise and distortion within social media datasets, the system gains improved reliability in recognizing harmful posts. Moreover, the work introduces an adapted metaheuristic optimization strategy designed to boost XGBoost classifier performance. Comparative evaluations conducted on authentic Twitter data confirm the efficiency of the method, with tuned classifiers reaching an accuracy of 91.86%, underscoring its promise for automated moderation and contributing to safer digital spaces. bookPart Springer, Cham 277 291 10.1007/978-3-032-28616-1_22 LNNS Lecture Notes in Networks and Systems: ICDPN 2025: International Conference on Data Processing and Networking, volume 2025