Pith. sign in

REVIEW 4 major objections 4 minor 47 references

This paper argues that each user's recommendation list should maximize their own expected matches given everyone else's lists, and proposes MODE, an algorithm that searches for such mutually optimal lists through best-response iteration.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

MODE iteratively recomputes each candidate's best recommendation list given others' lists, producing near-mutual optimality in direct effects with high expected matches and low runtime.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection MODE is a genuinely useful algorithm—correct DP and strong experiments—but the abstract overclaims match-rate superiority and the mutual-optimality guarantee is conditional on unproven convergence. the 4 major comments →

arxiv 2608.01731 v1 pith:FBUD2RBA submitted 2026-08-03 cs.IR

MODE: Mutual Optimality in Direct Effects of Reciprocal Recommendations in Matching Markets

Yoji Tomita This is my paper
classification cs.IR MSC 91B6891A10
keywords Reciprocal Recommender SystemsMatching MarketsMatching TheoryDirect EffectsMutual OptimalityBest-response dynamicsRanking probabilitiesSocial welfare
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper attacks a practical problem in reciprocal recommendation—dating apps and job boards where two sides must both accept. It argues that a user's satisfaction is mostly determined by the direct effect of their own recommendation list, and that fairness between users means no one's list should be sacrificed to help others. It defines mutual optimality in direct effects: every user's list maximizes their own expected number of matches, holding everyone else's list fixed. The proposed MODE method searches for such an outcome by repeatedly giving each user the best list against the others' current lists, using an exact recurrence for the rank that each user gets on each employer's list. The paper's experiments show that when this loop stabilizes—which it does in most tested cases—the result has near-zero per-user sub-optimality while remaining computationally cheap on markets of thousands of users.

Core claim

The central claim is that for a matching platform, the right optimization target is not only total expected matches but a per-user guarantee: given the recommendation lists of all other candidates, each candidate's own list should be the one that maximizes that candidate's expected matches (Definition 3.2, mutual optimality in direct effects). MODE computes this by best-response iteration. At each step it computes, for every candidate-employer pair, the probability that the candidate reaches a given rank on the employer's list (Proposition 3.3), converts that into a per-employer gain for the candidate, and sets the candidate's list to the top-K employers by gain. If the iteration converges,

What carries the argument

The ranking-probability recurrence (Proposition 3.3, Algorithm 1). For each employer, candidates are processed in the employer's preference order; the probability that candidate m occupies rank ℓ is built from the probability that the previous, more-preferred candidate occupies rank ℓ (didn't apply) or ℓ−1 (did apply). This turns the exponential subset sum in equation (6) into an exact dynamic program. MODE feeds these probabilities into a gain score g_ij = p_ij * sum_ℓ r_ij,ℓ w_j,ℓ q_ji; sorting employers by this score for each candidate is the best response, and repeating these best responses is the algorithm.

Load-bearing premise

The load-bearing premise is that repeatedly handing each user their best list will settle on a stable set of lists; if the loop keeps cycling, the output is just the best point in the cycle, not a mutually optimal recommendation, and the paper only reports empirical convergence.

What would settle it

Construct a small market (e.g., 3 candidates, 3 employers) where the reduced game has no pure Nash equilibrium—best responses cycle—and check whether MODE converges or returns a policy with positive sub-optimality. More broadly, measure cycle frequency and worst-case sub-optimality across many synthetic draws with correlated preferences; a nonzero sub-optimality on a sizable fraction of instances would directly contradict the 'nearly mutually optimal' claim.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If MODE converges, no user can increase their own expected number of matches by unilaterally changing their recommendation list; this is the definition of mutual optimality in direct effects.
  • The exact recurrence means sub-optimality of direct effects can be measured exactly per candidate, not estimated by Monte Carlo, so the reported near-zero sub-optimality is a precise quantity.
  • The per-iteration cost is polynomial in market size, and the paper reports seconds for n=100 where stochastic baselines take over ten hours, making the approach viable for large platforms.
  • In real-world 1000x1000 dating data, MODE beats the deterministic baselines by more than 10% in expected matches and approaches the heavier stochastic methods at smaller scale.
  • If the loop cycles, MODE falls back to the highest-social-welfare policy among those visited, so non-convergence degrades gracefully instead of failing.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The fixed-point guarantee is asymmetric: MODE optimizes only the proactive side's lists; applying the same best-response logic to the reactive side's ranked lists could yield a two-sided equilibrium notion and is a natural extension the paper does not develop.
  • Since mixed-strategy Nash equilibria are guaranteed to exist but pure ones are not, a smoothed or stochastic version of MODE could trade a little direct-effect optimality for a guaranteed stable outcome; the paper notes the existence result but does not pursue this direction.
  • The gain scores MODE computes are marginal match probabilities for each candidate-employer pair, so the same machinery could be plugged into global social-welfare optimization or fairness-constrained ranking, treating the paper's direct-effect objective as one component rather than the whole target.
  • The convergence rate of about 82% on default synthetic data and the low regret of cycles is reported for one preference model; on preference distributions with stronger correlations or larger markets, the iteration may cycle more often, and worst-case sub-optimality would be worth measuring.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper studies reciprocal recommendation in matching markets. It formalizes the optimality of direct effects of a recommendation list for an individual user, defines mutual optimality (a pure Nash equilibrium of the reduced game), and proposes MODE, an iterative best-response algorithm built on an efficient recurrence for rank probabilities (Prop. 3.3). Experiments on synthetic and real-world dating data compare MODE against Naive, Reciprocal, TU, ApproxSW, and DirectSW in terms of expected matches, direct-effect sub-optimality, computation time, and additional fairness metrics.

Significance. If the central claim held unconditionally, MODE would be a valuable practical method: it is deterministic, scales to large instances, and optimizes a well-motivated user-centric objective rather than a proxy. The rank-probability recurrence in Proposition 3.3 is a clean and useful computational tool. The experiments are carefully reported with confidence intervals and include a real-world dataset. However, the paper's headline claims—'computes mutually optimal recommendations' and 'higher expected number of matches in all cases'—are stronger than the evidence: mutual optimality is guaranteed only on convergence, which is not proven and is empirically partial, and the real-data 200x200 results contradict the 'higher in all cases' claim. The underlying algorithm and experiments are sound, but the presentation overstates the results.

major comments (4)
  1. [Algorithm 2, lines 9-14; Section 3.1.1; Appendix C.1] The mutual-optimality guarantee is conditional on the loop reaching a fixed point (line 10). The paper does not prove convergence, and Section 3.1.1 explicitly notes pure-strategy Nash equilibria need not exist. Appendix C.1 reports only 82% convergence in the default synthetic setting, with lower rates for other examination types and market sizes. In the non-convergence case, the algorithm returns the policy in the history with the highest social welfare (line 13), which has no mutual-optimality guarantee. The abstract's unqualified statement that MODE 'computes mutually optimal recommendations in direct effects' is therefore not supported by the paper's own definitions and evidence. The authors should either prove convergence for a meaningful class of instances or substantially temper the claim and explicitly state that the output is a heuristic with an empirically observed, but not gu
  2. [Section 4.2, Figure 2] The abstract and introduction claim that MODE 'enables a higher expected number of matches in all cases' or 'achieves a higher expected number of matches in all cases.' This is contradicted by the paper's own real-data results: in the 200x200 setting, MODE's expected matches are lower than those of ApproxSW and DirectSW (e.g., 1.466 vs 1.505 and 1.499 for Gender A; 1.867 vs 1.923 and 1.927 for Gender B in normalized units). The text in Section 4.2 correctly says MODE is 'slightly smaller' than the stochastic benchmarks, but the abstract and introduction are not aligned with this. The claim must be corrected to state the actual comparison, e.g., MODE is competitive with or superior to deterministic baselines and is substantially faster than stochastic methods.
  3. [Section 4.2; Eq. (9)] The paper's main selling point is mutual optimality in direct effects, but the real-world experiments report only expected matches, not the sub-optimality metric defined in Eq. (9). The claim in the conclusion that MODE 'yields little sub-optimality in direct effects' is therefore supported only on synthetic data. Moreover, even in the synthetic experiments, the sub-optimality numbers in Figure 1b are not broken down by whether MODE converged or cycled; Appendix C.1 shows convergence affects the guarantee but the paper does not report the distribution of sub-optimality for cycled runs. The authors should report direct-effect sub-optimality for the real-data experiments and, for synthetic settings, separate converged and cycled runs, so the reader can see the actual regret when the central guarantee is not attained.
  4. [Section 3.3 and Section 3.1.1] The algorithm's fixed point is a pure Nash equilibrium of the reduced game, but the paper does not provide any theoretical analysis of the best-response dynamics beyond the local 'if it converges, it is mutually optimal' observation. In particular, the relation to potential games or other sufficient conditions for convergence is not discussed, and the paper does not characterize when cycling can occur beyond the empirical report in Appendix C.1. This is not a fatal flaw for a heuristic paper, but a short discussion of why convergence might be expected (or a counterexample with substantial sub-optimality) would materially strengthen the central claim. At minimum, the open problem should be stated explicitly.
minor comments (4)
  1. [Eq. (7)] The notation in the lower-bound social-welfare expression is unclear: the subscript 'ℓ' in 'w_{j,ℓ}' is not defined in the expression, which seems to intend evaluation of the examination function at the expected rank. Please clarify or rewrite the formula.
  2. [Section 4.1] The text says 'the expected matches of each method increase linearly in n,' but Figure 1c shows a roughly linear trend only for n>=20; for n=10 the behavior is less clear. Please soften or qualify this statement.
  3. [Abstract and Introduction] The abstract says MODE 'computes mutually optimal recommendations in direct effects' without the '(nearly)' hedge that appears elsewhere in the paper. Please make the abstract match the body's more careful phrasing.
  4. [Algorithm 2 line 12] The cycle-detection condition 'sigma^t = sigma^s for some s <= t-2' can miss cycles of length greater than one if the same policy reappears after more than one step; also 't= T' is handled in the same branch. Please clarify that the cycle check covers all earlier policies, not just immediate predecessors, and state the memory requirement.

Circularity Check

1 steps flagged

MODE's mutual-optimality guarantee is conditional on convergence and is true by construction as a fixed point of its own best-response update; the independent social-welfare results are not circular.

specific steps
  1. self definitional [Section 3.3, Algorithm 2, lines 3-10; Section 4.1, Equation (9)]
    "If the updates terminate, that is, if σ^t = σ^{t−1}, then σ^t is mutually optimal in direct effects, and thus the algorithm returns it (line 10)."

    Mutual optimality in direct effects (Definition 3.2) is defined as every candidate's list maximizing own utility given others' lists. MODE's update (lines 3-8) constructs each candidate's list by selecting the top-K employers by g^t_{i,j}, which is exactly the argmax of U_i(·, σ^{t−1}) under the PBM utility model. Hence a fixed point of the update is a mutual-best-response profile by definition. The 'guarantee' is a restatement of the fixed-point condition, not an independent derivation. Likewise, the reported zero sub-optimality when converged follows immediately from the objective the algorithm is designed to minimize; it is not an empirical discovery. The paper is transparent about this, and it explicitly reports non-convergence/cycles in Appendix C.1, so the conditional statement is ac

full rationale

The core derivation chain is largely self-contained. Proposition 3.3 recursively computes ranking probabilities from the PBM equations (1)-(3) and is proven in Appendix A; it does not assume the conclusion. MODE's use of these probabilities to compute per-candidate gains and sort employers is a correct best-response construction. The mutual-optimality claim is explicitly conditional on convergence (Algorithm 2, line 10), and the paper acknowledges that pure-strategy equilibria need not exist (Section 3.1.1) and that MODE converges in only 82% of default synthetic instances (Appendix C.1). The conditional statement is true by construction—a fixed point of the best-response map is the definition of mutual optimality—so it is not an independent prediction, but the paper does not disguise it as one. The independent content is the social-welfare comparison: SW is computed from the same PBM but is not the objective MODE optimizes, and no constants are fitted to produce the reported SW gains. Self-citations to [39], [41], and [42] are used for baselines and related fairness work; they are not load-bearing for the algorithm's derivation. Overall, this is a minor near-circularity in the framing of the convergence guarantee, not a case where the central result reduces to its inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

MODE introduces no new entities and has no model parameters fitted to target outcomes. It relies on the standard PBM framework, independence of user actions, tie-free preferences, and exogenous preference estimates. The only hand-chosen knobs are the iteration bound and initial policy, which experiments show have limited effect. The direct-effects optimality concept is a new definition but not a free parameter.

free parameters (2)
  • Maximum iteration T
    Algorithm 2 stops after T iterations if no convergence; a termination hyperparameter, not fitted to match outcomes.
  • Initial policy sigma^0
    MODE can start from a uniformly random, random, or heuristic policy; Table 1 shows little effect on results, so it is not used to tune the target.
axioms (5)
  • domain assumption Position-based model (PBM) governs user examination and action (Equation 1).
    The utility formulation and MODE's gain computation rely on PBM with known examination probabilities v and w.
  • domain assumption Application events across candidates are independent.
    Proposition 3.3's recursion computes rank distributions using independent Bernoulli applications; this follows from the stochastic policy model but is an assumption about user behavior.
  • domain assumption Employer preference scores q have no ties.
    Assumed in Section 2.1 to define a unique ordering; if ties exist, the ranking probabilities and gains would need tie-breaking, which is not specified.
  • domain assumption Preferences p and q are given as inputs from a prior estimation step.
    MODE takes estimated preferences and examination probabilities as inputs; it does not address preference estimation, so the overall system's performance depends on these estimates.
  • domain assumption Expected number of matches is the appropriate utility and welfare measure.
    The paper equates user utility with expected matches and social welfare with total expected matches; this is the standard objective in the cited framework but is a modeling choice.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MODE: Mutual Optimality in Direct Effects of Reciprocal Recommendations in Matching Markets." pith.science (2026). https://pith.science/paper/FBUD2RBA

@misc{pith2026260801731,
  author       = {Pith},
  title        = {Pith review of: MODE: Mutual Optimality in Direct Effects of Reciprocal Recommendations in Matching Markets},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FBUD2RBA}},
  note         = {Machine review of arXiv:2608.01731}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Matching platforms such as job posting services and online dating platforms have become widely used over the past decade. For a matching platform to be successful, it is crucial to design appropriate reciprocal recommendation systems (RRSs) that consider the preferences of users on both sides (job candidates and employers) and prevent opportunities from being concentrated too heavily on a few popular users. However, prioritizing concentration mitigation too much can lead to recommending undesirable results to some individual users, resulting in their dissatisfaction. In this paper, we formulate the concept of ``optimality of direct effects'' of the recommendation list for an individual user, given the recommendations to other users. Furthermore, we propose a novel method, MODE, that computes mutually optimal recommendations in direct effects. Experiments with synthetic and real-world data demonstrate that MODE surpasses other existing methods in terms of mutual optimality of direct effects, exhibits faster processing speeds, and enables a higher expected number of matches.

Figures

Figures reproduced from arXiv: 2608.01731 by Yoji Tomita.

Figure 1
Figure 1. Figure 1: Results of experiments with synthetic data. The parameters are [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Results of experiments with real-world data. Expected matches are the mean of total matches in [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Results of additional synthetic experiments. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 41 canonical work pages

  1. [1]

    Aman Agarwal, Ivan Zaitsev, Xuanhui Wang, Cheng Li, Marc Najork, and Thorsten Joachims. 2019. Estimating position bias without intrusive interven- tions. InProceedings of the twelfth ACM international conference on web search and data mining. 474–482

    work page 2019

  2. [2]

    Nikolaos D Almalis, George A Tsihrintzis, and Nikolaos Karagiannis. 2014. A content based approach for recommending personnel for job positions. InIISA 2014, The 5th International Conference on Information, Intelligence, Systems and Applications. IEEE, 45–49

    work page 2014

  3. [3]

    Orestes Appel, Francisco Chiclana, Jenny Carter, and Hamido Fujita. 2017. Cross- ratio uninorms as an effective aggregation mechanism in sentiment analysis. Knowledge-Based Systems124 (2017), 16–22

    work page 2017

  4. [4]

    Gary S Becker. 1973. A theory of marriage: Part I.Journal of Political economy 81, 4 (1973), 813–846

    work page 1973

  5. [5]

    Garrett Birkhoff. 1946. Tres observaciones sobre el algebra lineal.Univ. Nac. Tucumán. Revista A.5 (1946), 147–151

    work page 1946

  6. [6]

    Kuan-Ming Chen, Yu-Wei Hsieh, and Ming-Jen Lin. 2023. Reducing recommen- dation inequality via two-sided matching: A field experiment of online dating. International Economic Review64, 3 (2023), 1201–1221

    work page 2023

  7. [7]

    Xi Chen, Xiaotie Deng, and Shang-Hua Teng. 2009. Settling the complexity of computing two-player Nash equilibria.J. ACM56, 3 (2009), 1–57

    work page 2009

  8. [8]

    Eugene Choo and Aloysius Siow. 2006. Who marries whom and why.Journal of political Economy114, 1 (2006), 175–201

    work page 2006

  9. [9]

    Gini Corrado. 1936. On the measure of concentration with special reference to income and statistics.Colorado College Publication, General Series208 (1936), 73–79

    work page 1936

  10. [10]

    Constantinos Daskalakis, Paul W Goldberg, and Christos H Papadimitriou. 2009. The complexity of computing a Nash equilibrium.Commun. ACM52, 2 (2009), 89–97

    work page 2009

  11. [11]

    Steven Diamond and Stephen Boyd. 2016. CVXPY: A Python-embedded modeling language for convex optimization.Journal of Machine Learning Research17, 83 (2016), 1–5

    2016

  12. [12]

    Virginie Do, Sam Corbett-Davies, Jamal Atif, and Nicolas Usunier. 2021. Two- sided fairness in rankings via Lorenz dominance.Advances in Neural Information Processing Systems34 (2021), 8596–8608

    work page 2021

  13. [13]

    Marguerite Frank and Philip Wolfe. 1956. An algorithm for quadratic program- ming.Naval research logistics quarterly3, 1-2 (1956), 95–110

    work page 1956

  14. [14]

    David Gale and Lloyd S Shapley. 1962. College admissions and the stability of marriage.The American mathematical monthly69, 1 (1962), 9–15

    work page 1962

  15. [15]

    Junling Hu and Michael P Wellman. 2003. Nash Q-learning for general-sum stochastic games.Journal of machine learning research4, Nov (2003), 1039–1069

    2003

  16. [16]

    Kalervo Järvelin and Jaana Kekäläinen. 2002. Cumulated gain-based evaluation of IR techniques.ACM Transactions on Information Systems (TOIS)20, 4 (2002), 422–446

    2002

  17. [17]

    Thorsten Joachims, Adith Swaminathan, and Tobias Schnabel. 2017. Unbiased learning-to-rank with biased feedback. InProceedings of the tenth ACM interna- tional conference on web search and data mining. 781–789

    2017

  18. [18]

    Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization tech- niques for recommender systems.Computer42, 8 (2009), 30–37

    2009

  19. [19]

    Carlton E Lemke and Joseph T Howson, Jr. 1964. Equilibrium points of bimatrix games.Journal of the Society for industrial and Applied Mathematics12, 2 (1964), 413–423

    work page 1964

  20. [20]

    Yunqi Li, Hanxiong Chen, Shuyuan Xu, Yingqiang Ge, Juntao Tan, Shuchang Liu, and Yongfeng Zhang. 2023. Fairness in recommendation: Foundations, methods, and applications.ACM Transactions on Intelligent Systems and Technology14, 5 (2023), 1–48

    work page 2023

  21. [21]

    2009.Matching theory

    László Lovász and Michael D Plummer. 2009.Matching theory. Vol. 367. American Mathematical Soc

    work page 2009

  22. [22]

    Tsunenori Mine, Tomoyuki Kakuta, and Akira Ono. 2013. Reciprocal recom- mendation for job matching with bidirectional feedback. In2013 Second IIAI International Conference on Advanced Applied Informatics. IEEE, 39–44

    work page 2013

  23. [23]

    John F Nash Jr. 1950. Equilibrium points in n-person games.Proceedings of the national academy of sciences36, 1 (1950), 48–49

    work page 1950

  24. [24]

    John F Nash Jr. 1951. Non-Cooperative Games.Annals of Mathematics54, 2 (1951), 286–295

    work page 1951

  25. [25]

    1944.Theory of Games and Economic Behavior

    John Von Neumann and Oskar Morgenstern. 1944.Theory of Games and Economic Behavior. Princeton University Press

    work page 1944

  26. [26]

    2025.Reciprocal Recommender Systems

    James Neve. 2025.Reciprocal Recommender Systems. Springer Nature

    work page 2025

  27. [27]

    James Neve and Ivan Palomares. 2019. Aggregation strategies in user-to-user re- ciprocal recommender systems. In2019 IEEE International Conference on Systems, Man and Cybernetics (SMC). IEEE, 4031–4036

    work page 2019

  28. [28]

    James Neve and Ivan Palomares. 2019. Latent factor models and aggregation oper- ators for collaborative filtering in reciprocal recommender systems. InProceedings of the 13th ACM conference on recommender systems. 219–227

    work page 2019

  29. [29]

    Brendan O’donoghue, Eric Chu, Neal Parikh, and Stephen Boyd. 2016. Conic op- timization via operator splitting and homogeneous self-dual embedding.Journal of Optimization Theory and Applications169, 3 (2016), 1042–1068

    work page 2016

  30. [30]

    Iván Palomares, Carlos Porcel, Luiz Pizzato, Ido Guy, and Enrique Herrera-Viedma

    work page

  31. [31]

    Luiz Pizzato, Tomasz Rej, Joshua Akehurst, Irena Koprinska, Kalina Yacef, and Judy Kay. 2013. Recommending people to people: the nature of reciprocal rec- ommenders with a case study in online dating.User Modeling and User-Adapted Interaction23, 5 (2013), 447–488

    work page 2013

  32. [32]

    Luiz Pizzato, Tomek Rej, Thomas Chung, Irena Koprinska, and Judy Kay. 2010. RECON: a reciprocal recommender for online dating. InProceedings of the fourth ACM conference on Recommender systems. 207–214

    work page 2010

  33. [33]

    Boyd A Potts, Hassan Khosravi, Carl Reidsema, Aneesha Bakharia, Mark Belono- goff, and Melanie Fleming. 2018. Reciprocal peer recommendation for learning purposes. InProceedings of the 8th international conference on learning analytics and knowledge. 226–235

    work page 2018

  34. [34]

    Alvin E Roth and Marilda Sotomayor. 1992. Two-sided matching.Handbook of game theory with economic applications1 (1992), 485–541

    work page 1992

  35. [35]

    Amar Saini, Florin Rusu, and Andrew Johnston. 2019. PrivateJobMatch: a privacy- oriented deferred multi-match recommender system for stable employment. In Proceedings of the 13th ACM Conference on Recommender Systems. 87–95

    work page 2019

  36. [36]

    Lloyd S Shapley and Martin Shubik. 1971. The assignment game I: The core. International Journal of game theory1, 1 (1971), 111–130

    work page 1971

  37. [37]

    Ashudeep Singh and Thorsten Joachims. 2018. Fairness of exposure in rankings. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 2219–2228

    2018

  38. [38]

    Yi Su, Magd Bayoumi, and Thorsten Joachims. 2022. Optimizing rankings for recommendation in matching markets. InProceedings of the ACM Web Conference

    work page 2022

  39. [39]

    Yoji Tomita, Riku Togashi, Yuriko Hashizume, and Naoto Ohsaka. 2023. Fast and examination-agnostic reciprocal recommendation in matching markets. In MODE: Mutual Optimality in Direct Effects of Reciprocal Recommendations in Matching Markets RecSys ’26, September 27-October 02, 2026, Minneapolis, MN, USA Proceedings of the 17th ACM Conference on Recommende...

    work page 2023

  40. [40]

    Yoji Tomita, Riku Togashi, and Daisuke Moriwaki. 2022. Matching theory-based recommender systems in online dating. InProceedings of the 16th ACM Conference on Recommender Systems. 538–541

    work page 2022

  41. [41]

    Yoji Tomita and Tomohiko Yokoyama. 2024. Fair reciprocal recommendation in matching markets. InProceedings of the 18th ACM Conference on Recommender Systems. 209–218

    work page 2024

  42. [42]

    Yoji Tomita and Tomohiko Yokoyama. 2026. Balancing Fairness and High Match Rates in Reciprocal Recommender Systems: A Nash Social Welfare Approach. ACM Transactions on Recommender Systems(2026)

    work page 2026

  43. [43]

    Bin Xia, Junjie Yin, Jian Xu, and Yun Li. 2019. WE-Rec: A fairness-aware reciprocal recommendation based on Walrasian equilibrium.Knowledge-Based Systems182 (2019), 104857

    work page 2019

  44. [44]

    Peng Xia, Benyuan Liu, Yizhou Sun, and Cindy Chen. 2015. Reciprocal recommen- dation system for online dating. InProceedings of the 2015 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining 2015. 234–241

    work page 2015

  45. [45]

    Hongtao Yu, Chaoran Liu, and Fuzhi Zhang. 2011. Reciprocal recommendation algorithm for the field of recruitment.Journal of Information & Computational Science8, 16 (2011), 4061–4068

    work page 2011

  46. [46]

    Lizi Zhang, Hui Fang, Wee Keong Ng, and Jie Zhang. 2011. Intrank: Interaction ranking-based trustworthy friend recommendation. In2011IEEE 10th International Conference on Trust, Security and Privacy in Computing and Communications. IEEE, 266–273. A Proof of Proposition 3.3 Fix an employer 𝑗∈J . A top candidate𝑖𝑗,1 for 𝑗 must be placed in the 1st rank if𝑖𝑗...

    work page 2011

  47. [2021]

    Reciprocal Recommender Systems: Analysis of state-of-art literature, chal- lenges and opportunities towards social recommendation.Information Fusion69 (2021), 103–127

    work page 2021

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.