Word Sense Disambiguation (WSD) is a challenging Natural Language Processing (NLP) task. In the case of the Arabic language, AWSD has its special challenges due to its unique features, such as being agglutinative, lacking diacritics, and having limited annotated resources. This study explores the impact of stop words, including particles, on the performance of supervised Arabic WSD models, a largely unaddressed aspect due to the scarcity of annotated resources. We constructed labeled context-gloss pairs for Arabic verbs, where the definition was extracted from the Arramooz Alwaseet dictionary, and fine-tuned the AraBERTv2 model for Arabic verb sense disambiguation. Five dataset versions were created to study the influence of particles and stop words. The results revealed that keeping stop words, including particles, significantly improved model performance, with an F1-score of 78.95%. The results provide valuable insights into improving WSD in the context of Arabic verbs.