Applying a Vector Search Method in Reference Service Question-Answer Retrieval Systems

We have developed a reference service question-answer retrieval system prototype based on dense vector search technology. When librarians receive reference questions, they can utilize reference service question-answer systems to determine if similar questions have been asked before and how they were resolved. Readers can also search the system to see if their questions have been answered. However, traditional reference service question-answer retrieval systems primarily rely on keyword-based searches, which may not capture questions with different phrasing but identical meaning. Vector search has emerged as a promising approach to address this limitation, and it has been widely researched and applied in various domains. Nevertheless, there is a lack of research on applying vector search in reference service question-answer retrieval systems. Thus, we conducted initial experiments in this area. Leveraging ElasticSearch vector search technology, which can efficiently search through large datasets, we adopted this technology in our study. For building our system, we collected question-answer pairs from the Taipei Public Library’s Online Reference Service Platform, one of Taiwan’s most prominent online reference services. We report the data set we used and the system architecture.