Pre-Trained Word Embeddings With Discrete Rat Swarm Optimizer to Improve Word Sense Disambiguation

Word sense disambiguation (WSD) remains a critical challenge in natural language processing (NLP), requiring precise identification of context-dependent word meanings. This article proposes a novel global optimization framework for WSD by reformulating it as a combinatorial optimization problem. We introduce a discrete variant of the rat swarm optimizer (RSO), a metaheuristic inspired by rat foraging behavior, to efficiently select optimal sense combinations across sentences. Unlike local disambiguation methods, our approach leverages phrase-level embeddings and WordNet gloss definitions within a unified semantic fitness function, enabling robust global sense selection. Experiments on SensEval and SemEval benchmarks demonstrate state-of-the-art performance, achieving accuracy improvements of 76.32% (SensEval2), 71.80% (SensEval3), 58.30% (SemEval2007), 77.17% (SemEval2013), and 74.40% (SemEval2015) surpassing previous methods. The effectiveness of the model shows its benefits for downstream NLP tasks such as machine translation, information retrieval, and document indexing. Future work will focus on hyperparameter optimization and multilingual adaptation to enhance scalability for low-resource languages.