| Open-domain question answering is one of the most popular NLP tasks in recent years. As performance enhancement for the retriever component of such systems is of utmost importance, the dense passage retrieval (DPR) model has emerged as the most effective approach. This research presents a multi-vectors dense retriever model as an improvement of the original DPR. The new model is evaluated and compared against the original DPR in terms of its effectiveness in retrieval. Results show that the multi-vectors dense retriever model is able to improve the performance of the original DPR by 3%. |
