eprintid: 10194856 rev_number: 7 eprint_status: archive userid: 699 dir: disk0/10/19/48/56 datestamp: 2024-07-19 12:28:04 lastmod: 2024-07-19 12:28:04 status_changed: 2024-07-19 12:28:04 type: proceedings_section metadata_visibility: show sword_depositor: 699 creators_name: Jin, J creators_name: He, Z creators_name: Yang, M creators_name: Zhang, W creators_name: Yu, Y creators_name: Wang, J creators_name: McAuley, J title: InfoRank: Unbiased Learning-to-Rank via Conditional Mutual Information Minimization ispublished: pub divisions: UCL divisions: B04 divisions: F48 note: This version is the author accepted manuscript. For information on re-use, please refer to the publisher’s terms and conditions. abstract: Ranking items regarding individual user interests is a core technique of multiple downstream tasks such as recommender systems. Learning such a personalized ranker typically relies on the implicit feedback from users' past click-through behaviors. However, collected feedback is biased toward previously highly-ranked items and directly learning from it would result in "rich-get-richer"phenomena. In this paper, we propose a simple yet sufficient unbiased learning-to-rank paradigm named InfoRank that aims to simultaneously address both position and popularity biases. We begin by consolidating the impacts of those biases into a single observation factor, thereby providing a unified approach to addressing bias-related issues. Subsequently, we minimize the mutual information between the observation estimation and the relevance estimation conditioned on the input features. By doing so, our relevance estimation can be proved to be free of bias. To implement InfoRank, we first incorporate an attention mechanism to capture latent correlations within user-item features, thereby generating estimations of observation and relevance. We then introduce a regularization term, grounded in conditional mutual information, to promote conditional independence between relevance estimation and observation estimation. Experimental evaluations conducted across three extensive recommendation and search datasets reveal that InfoRank learns more precise and unbiased ranking strategies. date: 2024-05-13 date_type: published publisher: Association for Computing Machinery (ACM) official_url: http://dx.doi.org/10.1145/3589334.3645356 oa_status: green full_text_type: other language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 2281728 doi: 10.1145/3589334.3645356 isbn_13: 9798400701719 lyricists_name: Wang, Jun lyricists_id: JWANG00 actors_name: Wang, Jun actors_id: JWANG00 actors_role: owner full_text_status: public pres_type: paper publication: WWW 2024 - Proceedings of the ACM Web Conference place_of_pub: Singapore, Singapore pagerange: 1350-1361 event_title: WWW '24: The ACM Web Conference 2024 book_title: WWW 2024 - Proceedings of the ACM Web Conference citation: Jin, J; He, Z; Yang, M; Zhang, W; Yu, Y; Wang, J; McAuley, J; (2024) InfoRank: Unbiased Learning-to-Rank via Conditional Mutual Information Minimization. In: WWW 2024 - Proceedings of the ACM Web Conference. (pp. pp. 1350-1361). Association for Computing Machinery (ACM): Singapore, Singapore. Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10194856/1/2401.12553v1.pdf