2026-05-04T20:59:54.136Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11578 2025-09-04T22:42:57Z 3

Optimized Classifying of User Review Sentiment via Modified Metaheuristic, Chapter in LNNS Lecture Notes in Networks and Systems: ICCIS 2024: Communication and Intelligent Systems, Springer, volume 1373 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11578 https://link.springer.com/chapter/10.1007/978-981-96-5729-2_12 L. Jovanovic S. Kozakijevic M. Antonijevic M. Milovanovic M. Mravik M. Zivkovic N. Bacanin As shopping shifts to the online domain, challenges arise. Consumers rely on online reviews, which play a crucial role in helping them make informed purchases. Customer feedback also serves as valuable market research. Manual analysis of reviews is inefficient. Artificial intelligence (AI) can streamline this process, yielding accurate results. Due to the evolving nature of language, traditional numerical machine learning models struggle with textual analysis, making natural language processing (NLP) a more effective approach. Bidirectional encoder representations from transformers (BERT) offers a method for encoding and analyzing text. This work tests how useful a proposed NLP-based approach using BERT is for evaluating whether customer feedback in product reviews is positive or negative. However, as performance of classifiers is coupled with proper parameter selection, a modified version of a optimization metaheuristics, specifically the firefly algorithm (FA), is introduced to help improve accuracy through parameter tuning. The best-performing models demonstrate admirable outcomes with sentiment classification accuracy being 0.868961. bookPart Springer, Singapore 177 190 10.1007/978-981-96-5729-2_12 LNNS Lecture Notes in Networks and Systems: ICCIS 2024: Communication and Intelligent Systems, volume 1373