Generative Retrieval as Multi-Vector Dense Retrieval - Shiguang Wu
Автор: SIGIR Virtual Forum
Загружено: 2025-03-31
Просмотров: 154
Описание:
Generative Retrieval as Multi-Vector Dense Retrieval: Towards the Understanding of Matching in Generative Retrieval
Abstract: Generative retrieval (GR) has emerged as a new paradigm in information retrieval. It aims to directly generate identifiers of relevant documents for a given query, unifying the indexing, retrieval, and ranking processes in retrieval systems into a single model. For such a fully neural indexing model, it is intriguing to explore the inner mechanism and its relationship with classical dense vector models. In this talk, I will share our perspective on understanding the matching framework of GR, showing that it shares the same framework for measuring query-document relevance as the multi-vector dense retrieval model. We hope this will help us understand the learned indexing structure of GR and inspire future research.
Bio: Shiguang Wu is a Graduate student at IRLab of Shandong University, supervised by Prof. Pengjie Ren and Prof. Zhaochun Ren. He obtained his bachelor's degree from Taishan College of Shandong University in 2023. His research interests lie in information retrieval, particularly generative retrieval and embedding models. Personal Website: https://furyton.github.io/
SIGIR VF Talk - Shiguang Wu
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
