.png){kind=icon}
返回
Deep neural ranking model using distributed smoothing
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2023.119913.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Information Retrieval (IR) provides access to unstructured data to satisfy user information needs within an extensive collection. In probabilistic IR models, the probability of an N-gram query is estimated by increasing the background probabilities related to lower-order N-grams. Inspired by the N-gram smoothing approach, we propose a distributed transformer-based deep neural ranking model. The proposed model, called DSDNN (Distributed Smoothing Deep Neural Network), distributes N-gram pattern matching into several building blocks with hierarchical convolutional structure. It captures deep semantic understanding using the context -aware learned language representation model. We conduct experiments on three datasets (Robust04, MS MARCO, and ANTIQUE) for three IR tasks (Ad-hoc information retrieval, document ranking, and community question answering). Our experimental results show the effectiveness of DSDNN in both IR tasks with respect to the comparative existing state-of-the-art neural ranking models.
Keyword:
Information retrieval
Deep neural ranking model
Smoothing
Word mismatch
Word embedding
期刊
IF:
7.5
论文数:
2.9W
被引数:
10.2W
机构
引用论文
.png){kind=icon}
查看更多内容