mBERT: A Query Refinement Model for Marathi Word Sense Disambiguation
Vivek A. Manwar, A. B. Manwar · 2025
Word Sense Disambiguation involves determining the appropriate meaning of a word within a given context. Natural language processing continues to face challenges related to ambiguity. Reducing these ambiguities for applications such as machine translation, data extraction, information retrieval, query response systems. In this paper, the proposed mBERT model for word disambiguation in Marathi query refinement show good performance with accuracy of 92.5%. This accuracy shows the model's proficiency in correctly interpret the intended meanings of ambiguous words within Marathi queries. With a precision of 91.2%, the model effectively minimizes false positives, ensuring that the refined queries remain relevant and meaningful and recall score is 90.8%, shows its capability to retrieve the most relevant information from the query context. The model's balanced performance is further highlighted by a high F1 score of 91.0%, which reflects a harmonious blend of precision and recall.