2026-06-08T14:51:10.992Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11958 2026-05-31T00:09:29Z 3

Tuning XGBoost Model for Twitter Sexism Detection Utilizing Gensim and Modified Particle Swarm Optimization Algorithm, Chapter in LNEE Lecture Notes in Electrical Engineering: ICMEET 2025: Evolution in Signal Processing and Telecommunication Networks, Springer, volume 1601 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11958 https://link.springer.com/chapter/10.1007/978-3-032-20544-5_10 E. Jeremic Markovic I. Kosta Z. Krsmanovic S. Anetic M. Zivkovic T. Zivkovic B. Radomirovic N. Bacanin The widespread use of social networks has drastically reshaped the way individuals connect and exchange information. Although these digital platforms have expanded the opportunities for interaction, they have simultaneously contributed to a troubling increase in misogynistic remarks and hostility-driven actions. Influential elements behind this trend include the concealment of user identities, unrestricted online self-expression, limited oversight of published material, as well as the persistence of discriminatory linguistic structures, nuanced textual signals, and contextual determinants. To confront this issue, the present work proposes a holistic framework that fuses artificial intelligence with sophisticated natural language processing methodologies and machine learning classifiers to uncover gender-biased discourse. Through the application of efficient text preprocessing procedures that minimize noise and alleviate bias in online communication, the system improves the precision of detecting prejudiced material. The investigation applies the Word2Vec Gensim embedding method and explores the potential of the XGBoost learning algorithm. Empirical findings validate the effectiveness of the proposed scheme in recognizing and categorizing sexist expressions on Twitter, thus contributing to automated moderation of digital content and encouraging a more equitable virtual space. In addition, the study introduces a customized metaheuristic optimization mechanism to further enhance classifier efficiency. Comparative evaluations on authentic datasets reveal that the refined model achieves a promising accuracy level of 95.37%. bookPart Springer, Cham 100 113 10.1007/978-3-032-20544-5_10 LNEE Lecture Notes in Electrical Engineering: ICMEET 2025: Proceedings of Tenth International Conference on Microelectronics Electromagnetics and Telecommunications, volume 1601