Zero-shot Information Retrieval

With advancements in Transformer models, Zero-shot Information Retrieval (IR) has seen remarkable progress. Document retrieval saves time for knowledge workers seeking relevant documents from a corpus. This study evaluates existing IR models in various Zero-shot scenarios, identifying limitations and potential improvements. It investigates data augmentation benefits for generative models in Zero-shot IR. The proposed retrieval system combines generative models, Bi-encoder, and hybrid search to understand knowledge transfer, develop domain-specific data augmentation, and assess BM25's impact on retrieval scores. Our contributions lie in highlighting areas for improvement and providing insights into model performance in diverse scenarios.