Contextual Query Expansion: Transforming Question Answering with an MLM-Based Approach Using A Transformer Model

The purpose of Question Answering (QA) frameworks is to provide precise answers to Natural Language Queries (NLQ) by extracting relevant information from large document collections. A key challenge is that users often struggle to formulate optimal queries, a limitation that prior methods have often failed to address by neglecting the valuable role of contextual information. To overcome this, this study introduces an MLM-based approach that takes advantage of contextual cues from document collection. By harnessing the context extraction capabilities of transformer models (such as BERT), expanded queries are automatically generated to better capture user intent. The proposed technique significantly improves the effectiveness of QA. The proposed MLM-generated expanded queries outperform query expansion methods based on WordNet, validating the integration of context clues as a promising development to refine the accuracy and relevance of QA responses.