Fine-Grained Distillation for Long Document Retrieval
Yucheng Zhou, Tao Shen, Xiubo Geng, Chongyang Tao, Jianbing Shen, Guodong Long, Can Xu, Daxin Jiang
摘要
Long document retrieval aims to fetch queryrelevant documents from a large-scale collection, where knowledge distillation has become de facto to improve a retriever by mimicking a heterogeneous yet powerful cross-encoder. However, in contrast to passages or sentences, retrieval on long documents suffers from the scope hypothesis that a long document may cover multiple topics. This maximizes their structure heterogeneity and poses a granularmismatch issue, leading to an inferior distillation efficacy. In this work, we propose a new learning framework, fine-grained distillation (FGD), for long-document retrievers. While preserving the conventional dense retrieval paradigm, it first produces global-consistent representations crossing different fine granularity and then applies multi-granular aligned distillation merely during training. In experiments, we evaluate our framework on two long-document retrieval benchmarks, which show state-of-the-art performance. * Work done during the internship at Microsoft. † Corresponding author. 1 Each entry of the collection can be any text granularity (e.g., sentence, passage, document) but we take 'document' to denote 'entry of collection' in this paper for clear writing. Document: … The European Renaissance was a time of massive economic and cultural growth following the stagnation of the Middle Ages. Beginning in Italy in the 14th century, the movement spread to all parts of the continent during the next 300 years. The outstanding cultural and artistic heritage of the Renaissance can still be seen today in many of the great cities of the period, including Florence and Venice in Italy, Bruges in Belgium and Toledo in Spain. It All Began in Florence Florence is the city where the Renaissance began, and where it reached its peak in the 15th and 16th centuries under the patronage of the powerful Medici family. Some of the greatest names in Renaissance art are associated with the city, including Leonardo da Vinci, Botticelli and Michelangelo. The poet Dante, the political theorist Machiavelli and the scientist Galileo also lived and worked in Florence. Buildings like the Pitti Palace, Uffizi Gallery and Florence Cathedral are among the masterpieces of Renaissance architecture. The Legacy of Venice To present-day tourists, Venice is renowned for its picturesque canals and its lack of motorized vehicles … Query1: 3 people who were important of the european renaissance time period Query2: what are the italian renaissance cities
问问这篇 Paper
智能体会读完全文。
Lune 把这篇 Paper 索引到了每一个公式,引用它的顶会 Paper 也一样。你提问,回答直接引用原文。
它引用的顶会 Paper18
- Deberta: decoding-Enhanced Bert with Disentangled AttentionPengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu ChenICLR 2021 · 被引用 3,729 次
- Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text RetrievalLee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang 等ICLR 2021 · 被引用 1,547 次
- Poly-encoders: Architectures and Pre-training Strategies for Fast and Accurate Multi-sentence ScoringSamuel Humeau, Kurt Shuster, Marie-Anne Lachaux, Jason WestonICLR 2020 · 被引用 316 次
- Vision-Language Pre-Training with Triple Contrastive LearningJinyu Yang, Jiali Duan, Son Tran, Yi Xu 等CVPR 2022 · 被引用 266 次
- Optimizing Dense Retrieval Model Training with Hard NegativesJingtao Zhan, Jiaxin Mao, Yiqun Liu, Jiafeng Guo 等SIGIR 2021 · 被引用 242 次
相关 Paper
- Towards Better Entity Linking with Multi-View Enhanced DistillationYi Liu, Yuan Tian, Jianxun Lian, Xinlong Wang 等ACL 2023 · 被引用 1 次
- PROD: Progressive Distillation for Dense RetrievalZhenghao Lin, Yeyun Gong, Xiao Liu, Hang Zhang 等WWW 2023 · 被引用 33 次
- Contextual Fine-to-Coarse Distillation for Coarse-grained Response Selection in Open-Domain ConversationsWei Chen, Yeyun Gong, Can Xu, Huang Hu 等ACL 2022
- Dual Learning with Dynamic Knowledge Distillation for Partially Relevant Video RetrievalJianfeng Dong, Minsong Zhang, Zheng Zhang, Xianke Chen 等ICCV 2023 · 被引用 35 次
- Grand: A Fast and Accurate Graph Retrieval Framework via Knowledge DistillationLin Lan, Pinghui Wang, Rui Shi, Tingqing Liu 等SIGIR 2024 · 被引用 3 次
