Parameterized Complexity of Fair Many-to-One Matchings

Hossein Shokouhi; Ramin Javadi

REVIEW 1 major objections 5 minor 24 references

Parameterized Complexity of Fair Many-to-One Matchings

T0 review · 1 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Generalized Fair Matching is fixed-parameter tractable for treewidth, neighborhood diversity, and tree-depth plus colors, yet W[1]-hard for feedback vertex number and path-width.

desk verdict Theorem 3's reduction is unsound: forced leaves make every constructed instance infeasible, so the (fvn,td) W[1]-hardness is unsupported; the FPT results look salvageable. read the letter →

arxiv 2411.18367 v1 pith:VBZON3WP submitted 2024-11-27 cs.CC

Ramin Javadi , Hossein Shokouhi This is my paper · ORCID

classification cs.CC MSC 05C7068Q2568R10

keywords fairmatchingmany-to-oneparameterizedcomplexityW[1]-hardnessfixed-parametertractabletree-depthpath-widthneighborhooddiversity

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Fair many-to-one matching asks for a left-perfect matching in a bipartite graph in which every right-side vertex receives a balanced multiset of colored left-side vertices. The paper establishes a parameterized complexity map for this problem: it is fixed-parameter tractable with respect to neighborhood diversity, to treewidth together with the maximum degree of the right side, and to tree-depth together with the number of colors; it is W[1]-hard with respect to feedback vertex number together with tree-depth even when left-side degrees are at most two, and with respect to path-width together with the number of colors. This matters because such matchings model fair assignment of people or items to resources, and the classification identifies which structural features of the input graph make the problem tractable in principle.

What carries the argument

The proof rests on two ILP formulations and on two gadget reductions. ILP1, carrying one pair of variables per vertex of V, generalizes the earlier |V|-parameterized FPT result. ILP2, with one binary variable per edge, has a constraint dual graph that the paper argues is a subgraph of the graph obtained by blowing up each V-vertex into a clique of size |C|+1; the dual-tree-depth ILP theorem then yields the (tree-depth, colors) FPT bound with runtime $O^*(2^{2^{td(|C|+1)}})$. The W[1]-hardness constructions encode Multicolored Clique into graphs of bounded feedback vertex number and tree-depth, and Unary Bin Packing into graphs of bounded path-width, using gadgets that pin the matching to a single choice per color class or per bin.

What would settle it

On small random bipartite graphs, construct ILP2 for a given coloring and compute the tree-depth of its dual graph; any instance where this tree-depth exceeds (|C|+1) times the input graph's tree-depth would refute the bound used in Theorem 8. Alternatively, explicitly search for a counterexample to the claimed subgraph containment of the dual graph in the blown-up graph.

Watch

Extended reading notes

Core claim

The central claim is a dichotomy for the Generalized Fair Matching problem. On the positive side, the paper proves FPT algorithms parameterized by (treewidth, maximum degree of V), by neighborhood diversity (which yields FPT for vertex cover and modular-width), by feedback edge number, and by (tree-depth, number of colors) through an ILP whose dual graph has bounded tree-depth. On the negative side, it proves W[1]-hardness for (feedback vertex number, tree-depth) even with Δ_U=2 via a reduction from Multicolored Clique, and for (path-width, number of colors, Δ_U) via a reduction from Unary Bin Packing. Together these results draw the algorithmic boundary of the problem in terms of structural graph parameters.

Load-bearing premise

The FPT result for tree-depth combined with the number of colors relies on the unproved structural claim that the integer program's dual constraint graph has tree-depth at most (|C|+1) times the tree-depth of the input graph; if that bound fails, the double-exponential runtime guarantee collapses.

Editorial extensions

If this is right

A correct FPT algorithm for (treewidth, Δ_V) means the problem is solvable in polynomial time on graphs of bounded treewidth whenever right-side degrees are bounded.
The neighborhood-diversity FPT result subsumes the earlier FPT bound in terms of |V| and transfers to vertex cover and modular-width parameters.
The (tree-depth, colors) result gives a double-exponential runtime that is nevertheless fixed-parameter tractable, so instances with small tree-depth and few colors are algorithmically accessible.
The W[1]-hardness for (feedback vertex number, tree-depth) rules out FPT algorithms for either parameter alone under standard complexity assumptions.
The path-width hardness shows that replacing tree-depth by path-width in the positive result is impossible, even when the number of colors and left-side degrees are bounded.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The tree-depth bound for the ILP dual graph in Theorem 8 is the most delicate step; if the asserted containment fails for some instance, the theorem would need a different structural argument to survive.
The hardness reductions use thresholds 0 and 1 on right-side vertices, so an FPT result for the uniform-threshold Fair Matching problem would not follow from these constructions.
Given the W[1]-hardness for (feedback vertex number, tree-depth), the open question of FPT for (feedback vertex number, number of colors) looks unlikely to have a positive answer.
The neighborhood-diversity FPT result suggests that a polynomial kernel for vertex cover may exist, which would strengthen the tractability picture.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The stress-test on Theorem 3 is correct, and it is a load-bearing problem. In the reduction, every vertex in U_{a,b} has degree one and its only neighbor is v_ab. U_{a,b} contains one vertex of every color except (b,0). Since L(v_ab)=0, any left-perfect matching gives v_ab at least n+1 vertices of some colors and zero of color (b,0), so the max-min gap is positive. The sufficiency construction in the proof explicitly matches all of U_{a,b} to v_ab, so it fails on its own terms. The reduction maps every Multicolored Clique instance to a no-instance. This is not a typo-level issue; it invalidates the W[1]-hardness with respect to (fvn, td) as stated.

The paper does have real value elsewhere. The FPT algorithm for neighborhood diversity genuinely generalizes the earlier |V|-FPT result of Boehmer and Koana, and the dynamic program for treewidth plus ΔV is recognizable and, as far as I checked, correct. The tree-depth plus number of colors result via dual tree-depth of an ILP is an elegant idea, and the claimed bound on the dual graph's tree-depth is plausible through the standard blow-up construction, although the paper only gestures at it. The feedback edge number DP in Theorem 5 also looks sound.

There are several smaller issues that a revision should clean up. In Theorem 8's ILP2, x_v and y_v are defined as maximum and minimum counts but constraint (9) has them backwards; that is almost certainly a variable-swap typo. In Theorem 7, the removal of K2 components is not correct when |C|=1 and L(v)=0: the component is then feasible, so the blanket claim 'the answer is no' is false. The later rounding of the residue r_c is also under-specified and can produce negative entries; a greedy allocation would fix it. None of these are as serious as the Theorem 3 flaw, but they should be addressed.

My bottom line: the paper deserves referee time—the topic is current, and the FPT side is a genuine contribution—but the referee should be directed at Theorem 3 first. I would not cite the W[1]-hardness result in its current form, and I would want to see a corrected reduction before recommending acceptance. If the authors can repair Theorem 3 (for instance by attaching a gadget to v_ab that supplies the missing color (b,0), or by changing L(v_ab)), the paper could be a good fit for a parameterized complexity venue.

Referee Report

1 major / 5 minor

Summary. The paper studies the parameterized complexity of Generalized Fair Matching, a left-perfect many-to-one matching problem with a fairness constraint on the color distribution at each right-hand vertex. The authors propose two W[1]-hardness results, one with respect to the feedback vertex number and tree-depth combined (Theorem 3) and one with respect to path-width, number of colors, and maximum degree of U combined (Theorem 4), as well as FPT results for feedback edge number (Theorem 5), treewidth plus maximum degree of V (Theorem 6), neighborhood diversity (Theorem 7), and tree-depth plus number of colors (Theorem 8). The hardness reductions are from Multicolored Clique and Unary Bin Packing; the FPT algorithms use dynamic programming, a neighborhood-diversity compression, and an ILP solved through a dual tree-depth bound.

Significance. If correct, the results would provide a fairly complete structural classification for a natural fair matching problem and would settle the parameterized status with respect to several important structural parameters. The claimed FPT results, especially the ILP-based tree-depth bound and the Unary Bin Packing reduction, are interesting and potentially valuable. However, the main W[1]-hardness contribution (Theorem 3) is invalid as written, and this undermines the paper's central classification claim. The paper contains no machine-checked proofs or reproducible code; the arguments are standard combinatorial constructions that require careful verification.

major comments (1)

[Section 3, Theorem 3] The reduction from Multicolored Clique is not sound. In the constructed instance, for every pair a,b with a≠b, each vertex of U_{a,b} has degree exactly 1 and is adjacent only to v_ab. Since the matching is left-perfect, every vertex of U_{a,b} must be matched to v_ab. The set U_{a,b} contains exactly one vertex of every color (c,i) with c∈[ℓ]\{b} and i∈[0,n], and it contains no vertex of color (b,0). Because v_ab receives no H[Ĉ] gadget, the proof states L(v_ab)=0. Hence in every left-perfect matching, v_ab has at least one matched vertex of every color c≠b and zero matched vertices of color (b,0), so max_c|M_c(v_ab)| - min_c|M_c(v_ab)| ≥ 1 > L(v_ab)=0. Thus every left-perfect matching is infeasible, and the reduction maps every Multicolored Clique instance to a no-instance. In particular, the sufficiency construction in the proof is not L-fair for v_ab, and the claimed W[1]-hardness of Generalized Fair Matching with respect to (fvn, td) is unsupported.

minor comments (5)

[Section 4, Theorem 8] The definition of the variables x_v and y_v is reversed: the text says they are respectively the maximum and minimum color counts, but the correct intended meaning (as in ILP1 and as required by constraints (8) and (9)) is x_v = min_c |M_c(v)| and y_v = max_c |M_c(v)|. With the printed definition, constraint (9) x_v ≤ count ≤ y_v is contradictory. Please correct the variable definitions.
[Section 4, Theorem 8] The tree-depth bound td(G̃) ≤ (|C|+1)td is asserted as 'clear' without proof. The bound is true: one can replace each V-vertex in an elimination forest of G with a path of |C|+1 vertices and attach each child subtree to the last vertex of the path, yielding an elimination forest of height at most (|C|+1)td. The manuscript should include this argument because the double-exponential runtime of the ILP solver depends on it.
[Section 5 (Concluding Remark)] The statement 'vc ≤ O(2^nd)' is false as written; for example, a complete bipartite graph K_{n,n} has neighborhood diversity 2 and vertex cover number n. The intended implication from FPT with respect to neighborhood diversity to FPT with respect to vertex cover number follows from the reverse inequality nd ≤ 2^{vc}. Please correct the direction and the justification.
[Section 4, Theorem 8] The runtime notation 'O*(22td(|C|+1))' is ambiguous. The proof gives O*((||A||∞+1)^{2^{td_D(A)}}) with td_D(A) ≤ (|C|+1)td, so the theorem should state a double-exponential bound such as O*(2^{2^{(|C|+1)td}}), with clear superscripts.
[Section 4, Theorem 7] In the construction of the sequences (r^1_c,...,r^s_c), the index k is not defined. It should be the largest integer with ∑_{j=1}^k ℓ_j ≤ r_c. Without this definition, the description is incomplete.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reasoning: all hardness reductions use external W[1]-hard source problems, and all FPT algorithms invoke independent ILP machinery from prior work.

full rationale

This paper does not exhibit circular derivation. The W[1]-hardness results in Theorems 3 and 4 are parameterized reductions from Multicolored Clique and Unary Bin Packing, respectively, both of which are established external source problems. The FPT results rely on external algorithmic theorems: Theorem 1 uses the ILP formulation and Lenstra-type result from Boehmer and Koana [4], and Theorem 8 invokes the dual-tree-depth ILP theorem of Eisenbrand et al. [9]. None of these citations is to the present authors' own prior work, and none smuggles in the target statement as an assumption. There are no fitted parameters renamed as predictions, no quantity defined in terms of the result it is supposed to predict, and no load-bearing self-citation chain. The reader's caution about Theorem 8 is legitimate but concerns a missing proof detail, namely the assertion that the dual graph is a subgraph of the blown-up graph and hence has tree-depth at most (|C|+1)·td(G); that is a correctness or completeness issue, not a circularity issue, because the bound is not assumed from the conclusion. Similarly, the skeptical concern about Theorem 3's construction would, if valid, be a soundness bug in the reduction, not a case of the conclusion being identical to the hypothesis. The manuscript is self-contained against external benchmarks in the sense that every upper bound invokes an external theorem with stated assumptions and every lower bound starts from an independent NP-hard or W[1]-hard problem. No circularity score above 0 is warranted.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claims rest on standard parameterized complexity machinery and on the correctness of the newly constructed reduction gadgets. No free parameters are fitted to data, and no new physical or mathematical entities are postulated. The main axioms are the external hardness results and ILP theorems cited in the paper, plus the standard assumption that tree decompositions are available or computable in FPT time.

assumptions (6)

standard math Multicolored Clique is W[1]-hard parameterized by clique size
Used as the source problem for the Theorem 3 reduction.
standard math Unary Bin Packing is W[1]-hard parameterized by number of bins
Used as the source problem for the Theorem 4 reduction.
standard math Integer linear programming with fixed dimension is FPT (Lenstra's theorem)
Underpins Theorem 1 and the neighborhood diversity result in Theorem 7 via the ILP1 formulation from Boehmer and Koana.
standard math ILP is FPT parameterized by dual tree-depth and maximum absolute coefficient (Theorem 2 from Eisenbrand et al.)
Underpins Theorem 8, using the dual tree-depth of the ILP2 coefficient matrix.
domain assumption A nice tree decomposition of width tw is given or computable in FPT time
Theorem 6 explicitly states 'assuming that nice tree decomposition of width tw is given'; this is standard in the area.
standard math In bipartite graphs, neighborhood diversity equals modular-width, and FPT with respect to neighborhood diversity implies FPT with respect to vertex cover number
Used to derive the corollaries after Theorem 7.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Parameterized Complexity of Fair Many-to-One Matchings." pith.science (2026). https://pith.science/paper/VBZON3WP

@misc{pith2026241118367,
  author       = {Pith},
  title        = {Pith review of: Parameterized Complexity of Fair Many-to-One Matchings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VBZON3WP}},
  note         = {Machine review of arXiv:2411.18367}
}

abstract

Given a bipartite graph $G=(U\cup V,E)$, a left-perfect many-to-one matching is a subset $M \subseteq E$ such that each vertex in $U$ is incident with exactly one edge in $M$. If $U$ is partitioned into some groups, the matching is called fair if for every $v\in V$, the difference between the number of vertices matched with $v$ in any two groups does not exceed a given threshold. In this paper, we investigate parameterized complexity of fair left-perfect many-to-one matching problem with respect to the structural parameters of the input graph. In particular, we prove that the problem is W[1]-hard with respect to the feedback vertex number, tree-depth and the maximum degree of $U$, combined. Also, it is W[1]-hard with respect to the path-width, the number of groups and the maximum degree of $U$, combined. In the positive side, we prove that the problem is FPT with respect to the treewidth and the maximum degree of $V$. Also, it is FPT with respect to the neighborhood diversity of the input graph (which implies being FPT with respect to vertex cover and modular-width). Finally, we prove that the problem is FPT with respect to the tree-depth and the number of groups.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 21 canonical work pages

[1]

Fair hierarchical clustering

Sara Ahmadian, Alessandro Epasto, Marina Knittel, Ravi Kumar, Mohammad Mahdian, Benjamin Moseley, Philip Pham, Sergei Vassilvitskii, and Yuyan Wang. Fair hierarchical clustering. In 34th Conference on Neural Information Processing Systems , 2020

work page 2020
[2]

Fair correlation clustering

Sara Ahmadian, Alessandro Epasto, Ravi Kumar, and Mohammad Mahdian. Fair correlation clustering. In International conference on artificial intelligence and statistics , pages 4195--4205. PMLR, 2020

work page 2020
[3]

The structural complexity landscape of finding balance-fair shortest paths

Matthias Bentert, Leon Kellerhals, and Rolf Niedermeier. The structural complexity landscape of finding balance-fair shortest paths. Theoretical Computer Science , 933:149--162, 2022

work page 2022
[4]

The complexity of finding fair many-to-one matchings

Niclas Boehmer and Tomohiro Koana. The complexity of finding fair many-to-one matchings. ACM Transactions on Algorithms , 20(2):1--37, 2024

work page 2024
[5]

A refined complexity analysis of fair districting over graphs

Niclas Boehmer, Tomohiro Koana, and Rolf Niedermeier. A refined complexity analysis of fair districting over graphs. Autonomous Agents and Multi-Agent Systems , 37(1):13, 2023

work page 2023
[6]

Matroids, matchings, and fairness

Flavio Chierichetti, Ravi Kumar, Silvio Lattanzi, and Sergei Vassilvtiskii. Matroids, matchings, and fairness. In The 22nd international conference on artificial intelligence and statistics , pages 2212--2220. PMLR, 2019

work page 2019
[7]

Parameterized algorithms , volume 5

Marek Cygan, Fedor V Fomin, ukasz Kowalik, Daniel Lokshtanov, D \'a niel Marx, Marcin Pilipczuk, Micha Pilipczuk, and Saket Saurabh. Parameterized algorithms , volume 5. Springer, 2015

2015
[8]

Fairness in graph mining: A survey

Yushun Dong, Jing Ma, Song Wang, Chen Chen, and Jundong Li. Fairness in graph mining: A survey. IEEE Transactions on Knowledge and Data Engineering , 35(10):10583--10602, 2023

work page 2023

Show all 24 references

[9]

An algorithmic theory of integer programming

Friedrich Eisenbrand, Christoph Hunkenschr \"o der, Kim-Manuel Klein, Martin Kouteck \`y , Asaf Levin, and Shmuel Onn. An algorithmic theory of integer programming. arXiv preprint arXiv:1904.01361 , 2019

1904 arXiv
[10]

On the parameterized complexity of multiple-interval graph problems

Michael R Fellows, Danny Hermelin, Frances Rosamond, and St \'e phane Vialette. On the parameterized complexity of multiple-interval graph problems. Theoretical computer science , 410(1):53--61, 2009

2009
[11]

Fair correlation clustering with global and local guarantees

Zachary Friggstad and Ramin Mousavi. Fair correlation clustering with global and local guarantees. In Algorithms and Data Structures: 17th International Symposium, WADS 2021, Virtual Event, August 9--11, 2021, Proceedings 17 , pages 414--427. Springer, 2021

2021
[12]

Modification-fair cluster editing

Vincent Froese, Leon Kellerhals, and Rolf Niedermeier. Modification-fair cluster editing. Social Network Analysis and Mining , 14(1):109, 2024

2024
[13]

Fixed-parameter algorithms for fair hitting set problems

Tanmay Inamdar, Lawqueen Kanesh, Madhumita Kundu, Nidhi Purohit, and Saket Saurabh. Fixed-parameter algorithms for fair hitting set problems. In 48th International Symposium on Mathematical Foundations of Computer Science (MFCS 2023) . Schloss-Dagstuhl-Leibniz Zentrum f \"u r ...

2023
[14]

Bin packing with fixed number of bins revisited

Klaus Jansen, Stefan Kratsch, D \'a niel Marx, and Ildik \'o Schlotter. Bin packing with fixed number of bins revisited. Journal of Computer and System Sciences , 79(1):39--49, 2013

2013
[15]

Parameterized complexity of fair feedback vertex set problem

Lawqueen Kanesh, Soumen Maity, Komal Muluk, and Saket Saurabh. Parameterized complexity of fair feedback vertex set problem. Theoretical Computer Science , 867:1--12, 2021

2021
[16]

Fair graph mining

Jian Kang and Hanghang Tong. Fair graph mining. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management , pages 4849--4852, 2021

2021
[17]

Adversarial graph embeddings for fair influence maximization over social networks

Moein Khajehnejad, Ahmad Asgharian Rezaei, Mahmoudreza Babaei, Jessica Hoffmann, Mahdi Jalili, and Adrian Weller. Adversarial graph embeddings for fair influence maximization over social networks. In Proceedings of the Twenty-Ninth International Conference on International Joi...

2021
[18]

Parameterized complexity of fair vertex evaluation problems

Du s an Knop, Tom \'a s Masa r \' k, and Tom \'a s Toufar. Parameterized complexity of fair vertex evaluation problems. In 44th International Symposium on Mathematical Foundations of Computer Science (MFCS 2019) . Schloss Dagstuhl--Leibniz-Zentrum f \"u r Informatik, 2019

2019
[19]

Fair edge deletion problems on tree-decomposable graphs and improper colorings

Petr Kolman, Bernard Lidick \`y , and Jean-S \'e bastien Sereni. Fair edge deletion problems on tree-decomposable graphs and improper colorings. 2010

2010
[20]

Structural Graph Parameters, Fine-Grained Complexity, and Approximation

Michael Lampis. Structural Graph Parameters, Fine-Grained Complexity, and Approximation . PhD thesis, Universit \'e Paris Dauphine, 2022

2022
[21]

Integer programming with a fixed number of variables

Hendrik W Lenstra Jr. Integer programming with a fixed number of variables. Mathematics of operations research , 8(4):538--548, 1983

1983
[22]

Fair edge deletion problems

Lishin Lin and Sartaj Sahni. Fair edge deletion problems. IEEE transactions on computers , 38(5):756--761, 1989

1989
[23]

Parameterized complexity of fair deletion problems

Tom \'a s Masa r \' k and Tom \'a s Toufar. Parameterized complexity of fair deletion problems. Discrete Applied Mathematics , 278:51--61, 2020

2020
[24]

Minimizing margin of victory for fair political and educational districting

Ana-Andreea Stoica, Abhijnan Chakraborty, Palash Dey, and Krishna P Gummadi. Minimizing margin of victory for fair political and educational districting. In Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems , pages 1305--1313, 2020

2020

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Fair hierarchical clustering

Sara Ahmadian, Alessandro Epasto, Marina Knittel, Ravi Kumar, Mohammad Mahdian, Benjamin Moseley, Philip Pham, Sergei Vassilvitskii, and Yuyan Wang. Fair hierarchical clustering. In 34th Conference on Neural Information Processing Systems , 2020

work page 2020

[2] [2]

Fair correlation clustering

Sara Ahmadian, Alessandro Epasto, Ravi Kumar, and Mohammad Mahdian. Fair correlation clustering. In International conference on artificial intelligence and statistics , pages 4195--4205. PMLR, 2020

work page 2020

[3] [3]

The structural complexity landscape of finding balance-fair shortest paths

Matthias Bentert, Leon Kellerhals, and Rolf Niedermeier. The structural complexity landscape of finding balance-fair shortest paths. Theoretical Computer Science , 933:149--162, 2022

work page 2022

[4] [4]

The complexity of finding fair many-to-one matchings

Niclas Boehmer and Tomohiro Koana. The complexity of finding fair many-to-one matchings. ACM Transactions on Algorithms , 20(2):1--37, 2024

work page 2024

[5] [5]

A refined complexity analysis of fair districting over graphs

Niclas Boehmer, Tomohiro Koana, and Rolf Niedermeier. A refined complexity analysis of fair districting over graphs. Autonomous Agents and Multi-Agent Systems , 37(1):13, 2023

work page 2023

[6] [6]

Matroids, matchings, and fairness

Flavio Chierichetti, Ravi Kumar, Silvio Lattanzi, and Sergei Vassilvtiskii. Matroids, matchings, and fairness. In The 22nd international conference on artificial intelligence and statistics , pages 2212--2220. PMLR, 2019

work page 2019

[7] [7]

Parameterized algorithms , volume 5

Marek Cygan, Fedor V Fomin, ukasz Kowalik, Daniel Lokshtanov, D \'a niel Marx, Marcin Pilipczuk, Micha Pilipczuk, and Saket Saurabh. Parameterized algorithms , volume 5. Springer, 2015

2015

[8] [8]

Fairness in graph mining: A survey

Yushun Dong, Jing Ma, Song Wang, Chen Chen, and Jundong Li. Fairness in graph mining: A survey. IEEE Transactions on Knowledge and Data Engineering , 35(10):10583--10602, 2023

work page 2023