Prompt-based Passage Retrieval Algorithm for Open-domain Question-answering

• 1、School of Computer Science(National Pilot Software Engineering School),Beijing University of Posts and Telecommunications, Beijing 100876

Abstract：The open domain question answering system aims to solve problems from people who may use informal natural language to ask questions. Without direct structural information, the question answering system must find answers from a large number of paragraphs. The open domain QA method typically adopts a two-stage paradigm: a retriever-reader paradigm, where the retriever selects candidate contexts containing the answer to the question from a paragraph corpus, and the reader checks the context to return the final answer range.Retriever is the key to improving the performance of question answering systems. However,retrieval models require a large number of supervised samples, in response to the above issues, this article proposes an article retrieval algorithm PPR based on prompt learning, which uses replaceable prompt templates to improve the ability of question answering retrieval tasks and bridge the gap between different data domains. Based on the characteristics of the input samples, multiple parts in the prompt template are replaced with different prompt words. PPR models the problem in each domain and the knowledge generalization of the article, while preserving the knowledge of each specific domain. PPR uses anunsupervisedlarge-scale corpus for pre training to enhance general retrieval capabilities. The comparative experimental results with other retrieval devices show that PPR improves performance in full data fine-tuning, few-shot sample, and zero-shot sample scenarios, demonstrating the effectiveness and innovation of this Algorithm .

Keywords： Artificial Intelligence Dense Retrieval Open-domain Question Answering Prompt Learning Few-shot Learning