Not All Relevance Scores are Equal: Efficient Uncertainty and Calibration Modeling for Deep Retrieval Models

被引:11
作者
Cohen, Daniel [1 ]
Mitra, Bhaskar [2 ]
Lesota, Oleg [3 ]
Rekabsaz, Navid [3 ,4 ]
Eickhoff, Carsten [1 ]
机构
[1] Brown Univ, Providence, RI 02912 USA
[2] Microsoft, Montreal, PQ, Canada
[3] Johannes Kepler Univ Linz, Linz, Austria
[4] Linz Inst Technol, AI Lab, Linz, Austria
来源
SIGIR '21 - PROCEEDINGS OF THE 44TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL | 2021年
关键词
uncertainty; neural networks; calibration; search;
D O I
10.1145/3404835.3462951
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
In any ranking system, the retrieval model outputs a single score for a document based on its belief on how relevant it is to a given search query. While retrieval models have continued to improve with the introduction of increasingly complex architectures, few works have investigated a retrieval model's belief in the score beyond the scope of a single value. We argue that capturing the model's uncertainty with respect to its own scoring of a document is a critical aspect of retrieval that allows for greater use of current models across new document distributions, collections, or even improving effectiveness for down-stream tasks. In this paper, we address this problem via an efficient Bayesian framework for retrieval models which captures the model's belief in the relevance score through a stochastic process while adding only negligible computational overhead. We evaluate this belief via a ranking based calibration metric showing that our approximate Bayesian framework significantly improves a retrieval model's ranking effectiveness through a risk aware reranking as well as its confidence calibration. Lastly, we demonstrate that this additional uncertainty information is actionable and reliable on down-stream tasks represented via cutoff prediction.
引用
收藏
页码:654 / 664
页数:11
相关论文
共 59 条
  • [1] [Anonymous], 2008, SIGIR, DOI [DOI 10.1145/1390334.1390446, 10.1145/, DOI 10.1145/1390334]
  • [2] [Anonymous], 1998, P SIGIR, DOI [DOI 10.1145/290941.291008, 10.1145/290941.291008]
  • [3] [Anonymous], 2010, ICML
  • [4] Aslam JA, 2007, LECT NOTES COMPUT SC, V4425, P198
  • [5] Chow Yinlam, 2015, Advances in Neural Information Processing Systems, V28, P1522
  • [6] Cross Domain Regularization for Neural Ranking Models using Adversarial Learning
    Cohen, Daniel
    Mitra, Bhaskar
    Hofmann, Katja
    Croft, W. Bruce
    [J]. ACM/SIGIR PROCEEDINGS 2018, 2018, : 1025 - 1028
  • [7] Craswell Nick., 2019, Overview of the TREC 2019 deep learning track
  • [8] Culpepper J. Shane, 2016, P 21 AUSTR S, P17
  • [9] Cummins R, 2011, PROCEEDINGS OF THE 34TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL (SIGIR'11), P1089
  • [10] Deeper Text Understanding for IR with Contextual Neural Language Modeling
    Dai, Zhuyun
    Callan, Jamie
    [J]. PROCEEDINGS OF THE 42ND INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL (SIGIR '19), 2019, : 985 - 988