Axiomatic Causal Interventions for Reverse Engineering Relevance Computation in Neural Retrieval Models

被引:0
作者
Chen, Catherine [1 ]
Merullo, Jack [1 ]
Eickhoff, Carsten [2 ]
机构
[1] Brown Univ, Providence, RI 02912 USA
[2] Univ Tubingen, Tubingen, Germany
来源
PROCEEDINGS OF THE 47TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL, SIGIR 2024 | 2024年
关键词
interpretability; neural ranking models; information retrieval axioms; search;
D O I
10.1145/3626772.3657841
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Neural models have demonstrated remarkable performance across diverse ranking tasks. However, the processes and internal mechanisms along which they determine relevance are still largely unknown. Existing approaches for analyzing neural ranker behavior with respect to IR properties rely either on assessing overall model behavior or employing probing methods that may offer an incomplete understanding of causal mechanisms. To provide a more granular understanding of internal model decision-making processes, we propose the use of causal interventions to reverse engineer neural rankers, and demonstrate how mechanistic interpretability methods can be used to isolate components satisfying term-frequency axioms within a ranking model. We identify a group of attention heads that detect duplicate tokens in earlier layers of the model, then communicate with downstream heads to compute overall document relevance. More generally, we propose that this style of mechanistic analysis opens up avenues for reverse engineering the processes neural retrieval models use to compute relevance. This work aims to initiate granular interpretability efforts that will not only benefit retrieval model development and training, but ultimately ensure safer deployment of these models.
引用
收藏
页码:1401 / 1410
页数:10
相关论文
共 37 条
[1]   Probing Classifiers: Promises, Shortcomings, and Advances [J].
Belinkov, Yonatan .
COMPUTATIONAL LINGUISTICS, 2022, 48 (01) :207-219
[2]   Analysis Methods in Neural Language Processing: A Survey [J].
Belinkov, Yonatan ;
Glass, James .
TRANSACTIONS OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, 2019, 7 :49-72
[3]  
Bruza P. D., 1994, SIGIR '94. Proceedings of the Seventeenth Annual International ACM-SIGIR Conference on Research and Development in Information Retrieval, P112
[4]   Diagnosing BERT with Retrieval Heuristics [J].
Camara, Arthur ;
Hauff, Claudia .
ADVANCES IN INFORMATION RETRIEVAL, ECIR 2020, PT I, 2020, 12035 :605-618
[5]   Axiomatically Regularized Pre-training for Ad hoc Search [J].
Chen, Jia ;
Liu, Yiqun ;
Fang, Yan ;
Mao, Jiaxin ;
Fang, Hui ;
Yang, Shenghao ;
Xie, Xiaohui ;
Zhang, Min ;
Ma, Shaoping .
PROCEEDINGS OF THE 45TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL (SIGIR '22), 2022, :1524-1534
[6]  
Choi Jeongwhan, 2022, arXiv
[7]  
Clark K., 2019, ARXIV
[8]  
Elhage N., 2021, A Mathematical Framework for Transformer Circuits
[9]   A Linguistic Study on Relevance Modeling in Information Retrieval [J].
Fan, Yixing ;
Guo, Jiafeng ;
Ma, Xinyu ;
Zhang, Ruqing ;
Lan, Yanyan ;
Cheng, Xueqi .
PROCEEDINGS OF THE WORLD WIDE WEB CONFERENCE 2021 (WWW 2021), 2021, :1053-1064
[10]   Match Your Words! A Study of Lexical Matching in Neural Information Retrieval [J].
Formal, Thibault ;
Piwowarski, Benjamin ;
Clinchant, Stephane .
ADVANCES IN INFORMATION RETRIEVAL, PT II, 2022, 13186 :120-127