Dynamic Injection of Entity Knowledge into Dense Retrievers
This addresses retrieval accuracy issues for queries with rare entities, offering an incremental improvement over existing dense retrievers.
The paper tackles the problem of dense retrievers struggling with queries involving less-frequent entities by proposing KPR, a BERT-based retriever enhanced with dynamic entity knowledge, which achieves state-of-the-art performance on two datasets and shows large gains on EntityQuestions.
Dense retrievers often struggle with queries involving less-frequent entities due to their limited entity knowledge. We propose the Knowledgeable Passage Retriever (KPR), a BERT-based retriever enhanced with a context-entity attention layer and dynamically updatable entity embeddings. This design enables KPR to incorporate external entity knowledge without retraining. Experiments on three datasets demonstrate that KPR consistently improves retrieval accuracy, with particularly large gains on the EntityQuestions dataset. When built on the off-the-shelf bge-base retriever, KPR achieves state-of-the-art performance among similarly sized models on two datasets. Models and code are released at https://github.com/knowledgeable-embedding/knowledgeable-embedding.