---
title: 'Refined (k,P)-Truss Model: Concepts & Algorithms'
url: https://www.emergentmind.com/topics/refined-k-p-truss-model
type: topic
---

# Refined (k,P)-Truss Model: Concepts & Algorithms

to=arxiv_search.search  ฝ่ายขายข่าว  պարզjson
{"query":"id:1906.00140 OR id:2411.06405 OR id:1205.6693 OR id:1906.12335 OR id:2102.07564 OR id:1806.05523","max_results":10,"sort_by":"relevance"}【อ่านข้อความเต็มjson
[{"arxiv_id":"1906.00140","title":"Fast Algorithm for K-Truss Discovery on Public-Private Graphs","authors":["Jiafeng Xie","Hong Cheng","Siqiang Luo","Ben Kao"],"published":"2019-06-01","abstract":"In public-private graphs, users share one public graph and have their own private graphs. A private graph consists of personal private contacts that only can be visible to its owner, e.g., hidden friend lists on Facebook and secret following on Sina Weibo. However, existing public-private analytic algorithms have not yet investigated the dense subgraph discovery of k-truss, where each edge is contained in at least k-2 triangles. This paper aims at finding k-truss efficiently in public-private graphs. The core of our solution is a novel algorithm to update k-truss with node insertions. We develop a classification-based hybrid strategy of node insertions and edge insertions to incrementally compute k-truss in public-private graphs. Extensive experiments validate the superiority of our proposed algorithms against state-of-the-art methods on real-world datasets."},{"arxiv_id":"2411.06405","title":"Parallel Higher-order Truss Decomposition","authors":["Xinmiao Liu","Shixuan Sun","Xin Huang","Xiaowei Yang","Jiacheng Xu","Ming Li","Yun Peng","Xuemin Lin"],"published":"2024-11-10","abstract":"The k-truss model is one of the most important models in cohesive subgraph analysis. The k-truss decomposition problem is to compute the trussness of each edge in a given graph, and has been extensively studied. However, the conventional k-truss model is difficult to characterize the fine-grained hierarchical structures in networks due to the neglect of high order information. To overcome the limitation, the higher-order truss model is proposed in the literature. However, the previous solutions only consider non-parallel scenarios. To fill the gap, in this paper, we conduct the first research to study the problem of parallel higher-order truss decomposition. Specifically, a parallel framework is first proposed. Moreover, several optimizations are further developed to accelerate the processing. Finally, experiments over 6 real-world networks are conducted to verify the performance of proposed methods."},{"arxiv_id":"1906.12335","title":"Critical Edge Identification: A K-Truss Based Model","authors":["Yilin Yang","Yizhou Sun","Jianzhong Li","Lu Qin","Rong-Hua Li"],"published":"2019-06-27","abstract":"In a social network, the strength of relationships between users can significantly affect the stability of the network. In this paper, we use the k-truss model to measure the stability of a social network. To identify critical connections, we propose a novel problem, named k-truss minimization. Given a social network G and a budget b, it aims to find b edges for deletion which can lead to the maximum number of edge breaks in the k-truss of G. We show that the problem is NP-hard. To accelerate the computation, novel pruning rules are developed to reduce the candidate size. In addition, we propose an upper bound based strategy to further reduce the searching space. Comprehensive experiments are conducted over real social networks to demonstrate the efficiency and effectiveness of the proposed techniques."},{"arxiv_id":"1205.6693","title":"Truss Decomposition in Massive Networks","authors":["Jia Wang","James Cheng"],"published":"2012-05-30","abstract":"The k-truss is a type of cohesive subgraphs proposed recently for the study of networks. While the problem of computing most cohesive subgraphs is NP-hard, there exists a polynomial time algorithm for computing k-truss. Compared with k-core which is also efficient to compute, k-truss represents the \"core\" of a k-core that keeps the key information of, while filtering out less important information from, the k-core. However, existing algorithms for computing k-truss are inefficient for handling today's massive networks. We first improve the existing in-memory algorithm for computing k-truss in networks of moderate size. Then, we propose two I/O-efficient algorithms to handle massive networks that cannot fit in main memory. Our experiments on real datasets verify the efficiency of our algorithms and the value of k-truss."},{"arxiv_id":"2102.07564","title":"STruD: Truss Decomposition of Simplicial Complexes","authors":["Francesco Bonchi","Dina Mahmoodian","Mirco Nanni","Giancarlo Ruffo","Michele U. Tomasi"],"published":"2021-02-15","abstract":"A simplicial complex is a generalization of a graph: a collection of n-ary relationships (instead of binary as the edges of a graph), named simplices. In this paper, we develop a new tool to study the structure of simplicial complexes: we generalize the graph notion of truss decomposition to complexes, and show that this more powerful representation gives rise to different properties compared to the graph-based one. This power, however, comes with important computational challenges derived from the combinatorial explosion caused by the downward closure property of complexes. Drawing upon ideas from itemset mining and similarity search, we design a memory-aware algorithm, dubbed STruD, which is able to efficiently compute the truss decomposition of a simplicial complex. STruD adapts its behavior to the amount of available memory by storing intermediate data in a compact way. We then devise a variant that computes directly the n simplices of maximum trussness. By applying STruD to several datasets, we prove its scalability, and provide an analysis of their structure. Finally, we show that the truss decomposition can be seen as a filtration, and as such it can be used to study the persistent homology of a dataset, a method for computing topological features at different spatial resolutions, prominent in Topological Data Analysis."},{"arxiv_id":"1806.05523","title":"Bounds and algorithms for graph trusses","authors":["Paul Burkhardt","Vance Faber","David G. Harris"],"published":"2018-06-14","abstract":"The $k$-truss, introduced by Cohen (2005), is a graph where every edge is incident to at least $k$ triangles. This is a relaxation of the clique. It has proved to be a useful tool in identifying cohesive subnetworks in a variety of real-world graphs. Despite its simplicity and its utility, the combinatorial and algorithmic aspects of trusses have not been thoroughly explored. We provide nearly-tight bounds on the edge counts of $k$-trusses. We also give two improved algorithms for finding trusses in large-scale graphs. First, we present a simplified and faster algorithm, based on approach discussed in Wang & Cheng (2012). Second, we present a theoretical algorithm based on fast matrix multiplication; this converts a triangle-generation algorithm of Bjorklund et al. (2014) into a dynamic data structure."}]
We have search results enough. Need write article with intro + 5-7 numbered h2 sections. Probably 6 sections. Need no prefatory meta. Must only article itself. Need comprehensive. Ground claims only in data block. Could integrate all papers. Focus on "Refined (k,P)-Truss Model", synthesizing from these papers as concept, noting no single canonical generic definition across sources. Need be neutral, distinguish interpretations. Need cite papers max once per section each.

Need structure likely:
Intro paragraph defining refined (k,P)-Truss as generalized family of truss models that augment conventional k with second parameter P controlling visibility/pattern/dimension/distance/robustness; not standardized.
Sections:
1. Standard k-truss foundation
2. Interpretations of parameter P
3. Major refined model families
4. Algorithmic principles
5. Structural properties and decomposition semantics
6. Applications, limitations, controversies/open problems

Need ensure factual claims verbatim from data. If infer, clearly mark. Need perhaps use table with paper / refinement / P interpretation. Could help. But article should be mostly paragraphs. Table with 3 columns max and brief cells. Detailed explanation in text before/after.

Need carefully phrase claims that some papers do not define explicit (k,P)-truss. Must mention the refined model is umbrella term rather than unified formalism. This is supported by data: 1906.00140 no explicit P; 2411.06405 not generic motif-based (k,P); 2102.07564 no explicit (k,P) but can map. So intro can say term is best understood as a family of generalizations rather than a single universally standardized definition. This is an inference synthesized from sources? Data says no single paper define generic. We can say "Across the cited literature, no single canonical formalism named '(k,P)-truss' is fixed; rather, the second parameter is instantiated in several ways." That's supported by combined data, okay.

Need mention standard definitions. Use both conventions careful: some papers use support >= k-2 (1205, 1906.00140, 1906.12335, 2411 with conventional), while 1806 uses support >= k under shifted convention. Need mention this notational divergence. Good controversy/misconception. Could say there are two conventions, one shifted by 2. Cite 1205 and 1806. Need be precise.

Need maybe discuss "largest subgraph" and maximality. 1205 defines largest subgraph. 1906.12335 defines maximal and non-trivial. 2411 defines maximal. Could mention variants. Must be careful. Maybe standard graph definition: largest subgraph such that every edge support at least k-2 [1205.6693; 1906.00140]. Also some formulations use maximal subgraph [1906.12335, 2411.06405]. But if citing both in one section, fine.

Need mention trussness: \phi(e), \tau(e), T_H(e), t(e,h), \mathrm{tr}_X(\sigma). Fine.

Need connect to public-private graphs, higher-order distance, simplicial complexes, critical-edge minimization. Need define "refined" as adding P which might represent user-private overlay \(G_u\), distance threshold \(h\), simplex dimension/support pattern \(q \to q+1\), or robustness-style constraint suggested by support slack and followers. But careful: 1906.12335 does not define (k,P), only suggests could inform refined model. So use suggestive wording.

Need include formulas from data exactly where necessary. Keep limited. For standard k-truss maybe
\[
\sup_H(e)=|\{\triangle uvw : ...\}|
\]
and \( \sup_H(e)\ge k-2\). Maybe maybe mention ordinary support in conventional and higher-order support \(sup_G(e,h)=|\triangle_G(e,h)|\). Also simplicial joist perhaps formula for trussness upper bound? Might be too much. We can include key formulas but concise.

Need mention algorithmic contributions and data details maybe experiments. But must ensure metrics exact and maybe table? We can include a small table summarizing families and papers. Let's do table with Model family / P interpretation / representative source. 3 cols. But must keep brief. Good.

Need maybe numbered sections titles with "## 1. ..." according to instruction. Six sections.

Let's draft content:
Intro paragraph define refined (k,P)-Truss Model as class of truss generalizations augmenting k-truss with second refinement parameter P encoding visibility context, distance threshold, support motif, simplex dimension, or robustness semantics. The sources show no single generic formalism; instead refine standard support-threshold decomposition while retaining edge- or simplex-centered trussness and nested subgraph/subcomplex structure [1205.6693 maybe and others].
Need cite in intro maybe [1205.6693], [2411.06405], [2102.07564], [1906.00140]. Avoid too many maybe 3. Okay.

Section 1 Standard foundation:
Define graph \(G=(V,E)\), support, k-truss, trussness, hierarchy \(E_{T_k} = \cup_{j \ge k}\Phi_j\). Note alternate convention in 1806 every edge incident to at least k triangles, equivalent to Cohen's (k+2)-truss. Mention k-truss is a (k-1)-core but not vice versa from 1205. Good.
Need maybe formula with support and condition. cite both [1205.6693] and [1806.05523].

Section 2 Semantics of P:
Paragraphs explaining P is not standardized. Table summarizing:
Personalized visibility context — \(P(u)=G_u\) or \(E_u\) — personalized public-private graph k-truss [1906.00140]
Distance threshold — \(P=h\) — (k,h)-truss [2411.06405]
Simplex support pattern/dimension — \(P \equiv q\) or \(q\to q+1\) / joist-based support [2102.07564]
Robustness/vulnerability cue — support slack or deletion persistence, suggestive not explicit — [1906.12335]
Could mention not arbitrary motif P in 2411, not explicit P in 1906.00140 and 2102.07564. Need mark suggestions as "This suggests". Good.

Section 3 Personalized and visibility-aware model:
Describe public-private graph model. \(G=(V,E)\), each user u private graph \(G_u=(V_u,E_u)\), \(E_u \cap E = \emptyset\). Personalized graph \(g_u = G \cup G_u = (V,E \cup E_u)\). Standard k-truss computed on \(g_u\). This is refined (k,P) where P is user-private overlay. Mention no multi-user access lattice, no at most P private edges, no "P users must see" semantics. Cite [1906.00140]. Could include update rules maybe mention node insertion, rules 1-3, (k,d)-neighborhood, Lemma 1 bounds. Need some algorithmic detail maybe in next section.

Section 4 Distance- and topology-refined models:
Subsection by paragraphs on (k,h)-truss: common h-neighbor, h-support \(sup_G(e,h)=|\triangle_G(e,h)|\), (k,h)-truss maximal subgraph with every edge support≥k-2, h-trussness \(t(e,h)\). When h=1 reduces to ordinary k-truss. H-index recurrence and convergence to \(t(e,h)-2\). Nesting property.
Then simplicial truss: joist, simplicial k-truss, trussness \(\mathrm{tr}_X(\sigma)\), uniqueness, containment, cross-dimensional monotonicity, lower and upper bounds, filtration and persistent homology. Make sure claims from 2102 paper. Cite [2411.06405], [2102.07564].

Section 5 Decomposition algorithms and computational frameworks:
Need cover classic peeling algorithms from 1205 and 1806, external-memory and practical. Then personalized node insertion from 1906.00140, Hybrid-PP classifier, specific features, performance metrics. Then parallel higher-order truss from 2411.06405. We need section not too huge. Can mention Algorithm 2 improved truss decomposition O(m^1.5) time O(m+n) space [1205.6693], practical O(m\bar\delta(G)) O(m) [1806.05523]. Personalized algorithms: Edge-PP, Vertex-PP, Hybrid-PP with Random Forest best overall and accuracies 84.4, 87.0, 86.5, 87.8, training time less than 1 second, speedups examples on SNAP, etc. Need precise. Node insertion vs edge insertion speedups exact examples. Could include because comprehensive. But maybe too many numbers. Since all claim appear verbatim in data. Fine. Could mention. H-index parallel framework with monotonicity and convergence and pruning lemma. cite accordingly maybe once each. Might need avoid too many citations maybe one at end of paragraph.

Section 6 Robustness, criticality, and boundaries of the concept:
Discuss k-truss minimization [1906.12335], NP-hard for k≥5, follower function monotonic not submodular, groups, upper bound. Use as robustness lens; note no explicit P but suggests resilience-based refinement. Then limitations and misconceptions: no generic P across literature; not always motif-based; some are edge-centered not clique-centered; public-private model binary visibility only; simplicial complex truss specific to joists, not arbitrary motifs; notation shifts by 2. Could mention standard truss may miss fine-grained hierarchy; higher-order addresses that. Cite papers once in section maybe [1906.12335], [2411.06405], [1906.00140], [2102.07564], [1806.05523]. Need at most once per section each.

Need ensure not overcite each sentence.

Need intro no heading. Then six `##` numbered sections.

Let's ensure every concrete claim from data. Need avoid invented statements like "encyclopedia article". Fine.

Need maybe use quotes for titles? If mention titles, double quotes. Could mention paper titles sparingly. But okay.

Need maybe not mention author groups unless from data? Search results include authors. Could reference authors. The task says reference specific papers, authors, and research groups when mentioned in data. Author names are from arxiv tool results, so okay. But not necessary. Could mention Wang and Cheng, Burkhardt, Faber, and Harris, Bonchi et al. That's fine because known from search results and data. Use citations. Maybe mention Huang et al. (2014) as prior edge-insertion framework because data says it. Okay.

Need perhaps use formula on higher-order H-index recurrence? Might be too much, but audience technical. Maybe include one recurrence. But keep limited. Could include important formula:
\( H^{(n)}sup(e(u,v)) = \mathcal H(\{\min(P^{(n)}(u,w),P^{(n)}(v,w)) \mid w \in \triangle_G((u,v),h)\})\). But maybe too complex. Yet for comprehensive article maybe yes in section 4 or 5. But maybe avoid because overkill and formula notation weird. Maybe mention in prose "an iterative H-index recurrence converges to \(t(e,h)-2\)" sufficient. Since formulas not essential. The instruction says mathematical notation when appears in original data. Fine to include limited formulas.

Need maybe mention in section 3 update rules:
Rules after node insertion/deletion: old edge with trussness above inserted node's max unaffected; any old edge changes by at most 1; define (k,d)-neighborhood and bounds \(k_{\text{low}}(e)\le \hat T(e)\le k_{\text{up}}(e)\) with gap at most 1. That's good, shows technical depth. Use exact notation from data. Cite [1906.00140].

Need maybe mention the exact object on which truss queries are answered: \(g_u\). Good.

Let's ensure no claims outside data. For example "no single canonical formalism" is synthesis from papers. It's okay if phrased carefully: "Across these formulations, the term is best understood as ..." maybe considered inference. Better say "Across the cited formulations, the second parameter is instantiated differently rather than fixed by a single generic definition." This is directly supported. Fine.

Let's draft with concise but comprehensive style.
The refined \((k,P)\)-Truss Model denotes a family of truss generalizations in which the classical cohesion threshold \(k\) is supplemented by a second refinement parameter \(P\). In the literature considered here, \(P\) is not fixed by a single canonical formalism. Instead, it is instantiated as a user-specific visibility overlay in public-private graphs, a hop parameter in higher-order edge trusses, or a simplex-dimension support pattern in simplicial complexes. What remains stable across these variants is the truss perspective itself: cohesion is enforced through a local support condition evaluated inside the retained subgraph or subcomplex, and decomposition yields nested levels of trussness rather than a single dense substructure [1205.6693; 1906.00140; 2411.06405; 2102.07564].

## 1. Classical truss foundation

The starting point is the standard graph-theoretic \(k\)-truss. For an undirected simple graph \(G=(V,E)\), edge support is the number of triangles containing an edge. In the formulation used by Wang and Cheng, and also in the public-private and critical-edge works, the support of \(e=(u,v)\) in a subgraph \(H\) is
\[
\sup_H(e)=|\{\triangle uvw : (u,v),(u,w),(v,w)\in E(H)\}|,
\]
and a \(k\)-truss is the largest subgraph \(H\subseteq G\) such that every retained edge satisfies
\[
\sup_H(e)\ge k-2.
\]
Support is therefore subgraph-relative: it is measured inside the current residual structure, not in the original graph [1205.6693; 1906.00140].

Edge trussness indexes this hierarchy. In one notation, the trussness of an edge is
\[
\phi(e)=\max\{k: e\in E_{T_k}\},
\]
and the \(k\)-class is \(\Phi_k=\{e:e\in E_G,\ \phi(e)=k\}\), with
\[
E_{T_k}=\bigcup_{j\ge k}\Phi_j.
\]
This yields the standard nested decomposition of edges by triangle-supported cohesion [1205.6693].

A well-known notational divergence must be kept explicit. Burkhardt, Faber, and Harris define a \(k\)-truss as a graph in which every edge is incident to at least \(k\) triangles, rather than at least \(k-2\) triangles. They note that this is a shifted convention: their \(k\)-truss corresponds to Cohen’s \((k+2)\)-truss [1806.05523]. This convention shift is a recurrent source of confusion in the truss literature.

The classical model is also positioned between \(k\)-core and clique. Wang and Cheng state that “a \(k\)-truss is a \((k-1)\)-core but not vice versa,” while Burkhardt, Faber, and Harris emphasize that a \((k+2)\)-clique is a \(k\)-truss under their convention [1205.6693; 1806.05523]. The refined \((k,P)\)-Truss idea preserves this intermediate role while altering what counts as support.

## 2. Interpretations of the refinement parameter \(P\)

Across the cited work, the second parameter does not denote one universally accepted object. It is more accurate to treat \((k,P)\)-truss as an umbrella reading over several refinement dimensions.

| Refinement family | Interpretation of \(P\) | Representative source |
|---|---|---|
| Personalized truss | user-private overlay \(G_u\) or \(E_u\) | [1906.00140] |
| Higher-order edge truss | distance threshold \(h\) in \((k,h)\)-truss | [2411.06405] |
| Simplicial truss | simplex dimension or support pattern \(q\to q+1\) | [2102.07564] |
| Robustness-oriented reading | support slack or deletion persistence | [1906.12335] |

In the public-private setting, \(P\) is most naturally interpreted as the user-specific private graph \(G_u\) or private edge set \(E_u\). The truss is then simply the standard \(k\)-truss computed on the personalized accessible graph \(g_u=G\cup G_u\). The paper is explicit that it does not introduce a parameter named \(P\), but its model provides a direct personalized interpretation of such a parameter [1906.00140].

In the higher-order edge-truss setting, the literature does define a direct refinement, but as \((k,h)\)-truss rather than generic \((k,P)\)-truss. Here \(P\) is best read as the distance-based support pattern induced by common \(h\)-neighbors and \(h\)-hop connectivity. This is not a motif-truss over arbitrary patterns, but a distance-refined edge-truss model [2411.06405].

In simplicial complexes, the analogous refinement is even more structurally specific. The generalized truss is based on joists, meaning the set of all cofaces of a simplex, and support is induced dimension-by-dimension from \(q\)-simplices to \((q+1)\)-simplices. This suggests a refined \((k,P)\)-reading in which \(P\) is the dimension \(q\), the relation \(q\to q+1\), or the simplex-boundary support motif itself [2102.07564].

A robustness-style reading of \(P\) is only suggestive rather than explicit. The critical-edge work does not define \((k,P)\)-truss, but it develops notions such as support slack, followers, and budgeted deletion cascades. This suggests a possible refinement in which \(P\) captures persistence or resilience under support loss, although that interpretation is not formalized in the paper [1906.12335].

## 3. Personalized visibility-aware trusses on public-private graphs

The public-private graph model gives the clearest personalized reading of a refined \((k,P)\)-truss. The base graph is a simple undirected public graph
\[
G=(V,E),
\]
with neighborhood \(N(v)=\{u\in V:(v,u)\in E\}\) and degree \(d(v)=|N(v)|\). For each vertex \(u\in V\), there is an associated private graph
\[
G_u=(V_u,E_u),
\]
where \(V_u\subseteq V\) and \(E_u\cap E=\emptyset\). The visibility semantics are binary: the public graph is visible to everyone, while \(G_u\) is visible only to user \(u\). The personalized public-private graph is therefore
\[
g_u=G\cup G_u=(V,E\cup E_u),
\]
and the query problem is to find the \(k\)-truss in \(g_u\) [1906.00140].

Under a refined \((k,P)\)-interpretation, the role of \(P\) is played by the user-conditioned visibility context. The truss definition itself is unchanged; what changes is the graph on which support is evaluated. This matters because two users sharing the same public network may obtain different trusses solely because their private overlays differ.

The paper develops this idea algorithmically through incremental maintenance. A key observation is that a personalized query can be processed either by repeated edge insertion of private edges or by treating the user-specific augmentation as a node-insertion event. This leads to the distinction between **Edge-PP**, which adds private edges one by one, and **Vertex-PP**, which removes a user vertex together with its public edges and then reinserts it with all public and private incident edges [1906.00140].

The node-insertion theory is precise. If a new node \(v\) is inserted and its incident structure has updated trussness level \(l\), then every old edge with trussness at least \(l\) is unaffected. Under node deletion, every old edge with trussness strictly greater than the deleted node’s maximum incident trussness is unaffected. More generally, for every unaffected old edge outside the incident set of \(v\),
\[
|\hat T(e)-T(e)|\le 1.
\]
The paper explains this by the fact that a single node insertion or deletion can change an edge’s support by at most one triangle involving that node [1906.00140].

To tighten updates on new incident edges, the paper introduces the \((k,d)\)-neighborhood of a vertex \(v\), denoted \(G_{k,d}\), as the maximal subgraph \(H\subseteq G[N(v)]\) such that all its edges satisfy \(T_G(e)\ge k\) and all its vertices have degree at least \(d\). For each new edge \(e=(v,w)\), the updated trussness satisfies
\[
k_{\text{low}}(e)\le \hat T(e)\le k_{\text{up}}(e),
\qquad
|k_{\text{up}}(e)-k_{\text{low}}(e)|\le 1.
\]
This near-identification of new-edge trussness enables localized maintenance rather than recomputation [1906.00140].

The same work also makes the operational scope of the model explicit. It does not define a multi-user access-control lattice, privacy budgets, or semantics such as “at most \(P\) private edges” or “visible to \(P\) users.” Its refinement is narrower: one public graph, one private overlay per user, and truss queries answered in that user’s personalized view [1906.00140].

## 4. Distance- and topology-refined truss models

A distinct refinement axis replaces triangle support by broader higher-order support. In the \((k,h)\)-truss model, for an edge \(e=(u,v)\), a vertex \(w\) is a common \(h\)-neighbor if it is within \(h\) hops of both \(u\) and \(v\). The set of common \(h\)-neighbors is denoted \(\triangle_G(e,h)\), and the higher-order support of \(e\) is
\[
sup_G(e,h)=|\triangle_G(e,h)|.
\]
A subgraph \(S\subseteq G\) is a \((k,h)\)-truss if every edge in \(S\) has \(h\)-support at least \(k-2\) in \(S\), and \(S\) is maximal. The edge quantity
\[
t(e,h)=\max\{k\mid e \text{ is contained in a } (k,h)\text{-truss}\}
\]
is the \(h\)-trussness of \(e\) [2411.06405].

This model reduces exactly to ordinary truss when \(h=1\): common \(1\)-neighbors are ordinary common neighbors, \(sup_G(e,1)=sup_G(e)\), and \((k,1)\)-truss coincides with standard \(k\)-truss. For \(h>1\), support is no longer confined to triangles, but is mediated by vertices jointly close to both endpoints. The paper states that the conventional \(k\)-truss model is difficult to characterize fine-grained hierarchical structures because it neglects high-order information, and it shows on a toy graph that ordinary truss yields only \(2\)-truss and \(3\)-truss, whereas \((k,2)\)-truss yields \((4,2)\)-truss, \((5,2)\)-truss, and \((6,2)\)-truss [2411.06405].

The higher-order decomposition also admits an iterative H-index characterization. With initialization \(H^{(0)}sup(e)=sup_G(e,h)\), the paper defines a recurrence based on common \(h\)-neighbor witnesses and \(h\)-hop reachable path keys, proves monotonicity
\[
H^{(n)}sup(e)\ge H^{(n+1)}sup(e),
\]
and proves convergence
\[
\lim_{n\to\infty} H^{(n)}sup(e)=t(e,h)-2.
\]
This converts higher-order trussness from a sequential peeling notion into a fixed-point computation suitable for parallelization [2411.06405].

A more radical topological refinement appears in simplicial complexes. There the support primitive is not the triangle but the **joist**: for a simplex \(\sigma^{(q)}\), the joist \(J_{\sigma^{(q)}}\) is the set of all its cofaces. A simplicial \(k\)-truss is a maximal set of simplices \(T_k\subseteq K\) of dimension greater than \(0\) such that each simplex belongs to at least \(k\) joists entirely contained in \(T_k\). The maximal \(k\) such that \(\sigma^{(q)}\in T_k\) is the simplicial trussness \(\mathrm{tr}_X(\sigma^{(q)})\) [2102.07564].

This generalized model preserves decomposition structure but changes the support geometry completely. The paper proves uniqueness of the simplicial \(k\)-truss, containment
\[
T_{k+1}\subseteq T_k,
\]
and cross-dimensional anti-monotonicity:
\[
\sigma^{(q)}\subset \sigma^{(q+1)} \Rightarrow \mathrm{tr}_X(\sigma^{(q)})\ge \mathrm{tr}_X(\sigma^{(q+1)}).
\]
It also gives a lower bound based on the largest containing simplex and an upper bound based on the number of joists containing a simplex [2102.07564].

This suggests that a refined \((k,P)\)-truss may be interpreted not merely as an edge-based graph notion but as a dimension-aware support framework. Under that reading, \(P\) is naturally identified with \(q\), with the support relation \(q\to q+1\), or with the simplex-boundary motif \(\partial\Delta^{q+1}\). The paper itself does not introduce generic \((k,P)\)-notation, but it provides a precise higher-order truss formalism of that type [2102.07564].

## 5. Decomposition algorithms and computational architecture

The computational core of truss analysis remains peeling and support maintenance. Wang and Cheng give an improved in-memory decomposition that computes all \(k\)-classes directly. It computes support for each edge, sorts edges in ascending support, repeatedly removes the edge of lowest support, decrements supports of triangle-adjacent edges, and proves that all \(k\)-trusses for \(k\ge 3\) can be computed in \(O(m^{1.5})\) time using \(O(m+n)\) space [1205.6693].

The same paper develops two external-memory strategies for massive graphs: a bottom-up method that extracts \(k\)-classes using neighborhood subgraphs and safe lower bounds \(\varphi(e)\), and a top-down method for only the highest truss classes using upper bounds \(\psi(e)\). The neighborhood subgraph
\[
NS(U)
\]
contains internal vertices \(U\), all their neighbors, and all edges incident from \(U\), allowing exact support computation for internal edges. This architecture is central for out-of-core truss processing [1205.6693].

Burkhardt, Faber, and Harris present a simplified practical algorithm with time \(O(m\bar\delta(G))\) and memory \(O(m)\), where
\[
\bar\delta(G)=\frac{1}{m}\sum_{\text{edges }e=(u,v)}\min(d(u),d(v)).
\]
They also prove \(\bar\delta(G)\le 2\alpha(G)\le O(\sqrt m)\), yielding \(O(m^{3/2})\) in the worst case. Their second algorithm is a truncated decomposition method based on dynamic triangle maintenance, randomization, and fast matrix multiplication [1806.05523].

In personalized public-private graphs, the main algorithmic novelty is incremental truss discovery via node insertion. The paper defines **Edge-PP**, **Vertex-PP**, and a classification-based **Hybrid-PP** that chooses the faster strategy per query user. The feature vector includes the public degree of the user, private degree, numbers of triangles containing the user in the public graph, private graph, and pp-graph, sums and maxima of trussness on public incident edges, and summaries of the lower and upper bounds produced by the node-insertion bound algorithm. Five classifiers were tested, and Random Forest performed best overall, with accuracies on PP-DBLP of \(84.4\%\), \(87.0\%\), \(86.5\%\), and \(87.8\%\); training time was reported as less than 1 second [1906.00140].

The same study reports that on SNAP graphs, node insertion dramatically outperformed repeated edge insertion, with example speedups of \(3.20\times\) on DBLP, \(7.99\times\) on AstroPh, \(10.96\times\) on EmailEuAll, \(39.10\times\) on Wikivote, \(38.67\times\) on LiveJournal, and \(49.76\times\) on WikiTalk. On PP-DBLP, Hybrid-PP and Vertex-PP outperformed Edge-PP and recomputation across varying query-node degree and varying \(k\in\{5,7,9,11,13,15\}\) [1906.00140].

For higher-order \((k,h)\)-truss, the decomposition problem is to compute \(t(e,h)\) for every edge. The baseline is still peeling, but the 2024 work studies the first parallel higher-order truss decomposition framework. Its fixed-point formulation, monotonicity theorem, convergence theorem, asynchronous updating argument, and pruning lemma collectively replace serial edge peeling by parallel iterative local updates [2411.06405].

In simplicial complexes, STruD generalizes peeling to a setting in which candidate support structures are joists rather than triangles. It proceeds bottom-up in simplex dimension, extending simplices, finding joists by an inverted-index strategy, and then peeling by current trussness estimate. The algorithm is memory-aware and shifts to out-of-core candidate storage when joist structures exceed RAM. The worst-case cost reflects the combinatorial explosion caused by downward closure; for a clique complex over \(m\) vertices, the total cost of the main procedure is upper bounded by \(O(4^m)\) [2102.07564].

## 6. Robustness, criticality, and conceptual boundaries

A refined \((k,P)\)-truss can also be read through the lens of structural vulnerability. In the \(k\)-truss minimization problem, given a graph \(G\) and budget \(b\), one seeks a set \(B^*\) of \(b\) edges such that \(|T_k(B^*)|\) is minimized, equivalently
\[
B^*=\argmax_{B\subseteq E(G),\ |B|=b}|F(B,T_k)|,
\]
where \(F(B,T_k)\) is the follower set, namely the edges removed from the \(k\)-truss due to deletion of \(B\). The paper proves that for \(k\ge 5\) the problem is NP-hard, and that \(f(B)=|F(B,T_k)|\) is monotonic but not submodular [1906.12335].

The associated structural notions are useful even outside the exact minimization problem. The fragile-edge set
\[
P=\{e\mid sup(e,T_k)=k-2\}
\]
identifies support-critical edges. A candidate edge can trigger followers only if it is triangle-adjacent to some edge in this set. The paper then develops \(k\)-support groups, \(k\)-truss groups, follower dominance, and an upper bound
\[
|F(e,T_k)|\le \sum_{i=1}^{\theta}|E(g_i)|
\]
when \(e\) is triangle-adjacent to \(k\)-truss groups \(g_1,\dots,g_\theta\). These ideas do not define an explicit \(P\), but they suggest a refinement axis based on resilience margin or permissible support loss [1906.12335].

Several conceptual boundaries emerge from the literature. First, a refined truss need not be motif-parametrized in a generic sense. The \((k,h)\)-truss is distance-refined rather than arbitrary-motif-based, and its decomposed object remains an edge rather than a higher-order clique [2411.06405]. Second, the public-private formulation is privacy-conditioned but limited to one public graph plus one user-specific private overlay with binary visibility semantics [1906.00140]. Third, the simplicial generalization is not a free-form higher-order support system: its support object is specifically the joist, i.e., the boundary relation induced by a \((q+1)\)-simplex [2102.07564].

The literature also warns against overly dense intuitions. Burkhardt, Faber, and Harris prove nearly-tight bounds showing that connected \(k\)-trusses can be sparse, with minimum edge count between
\[
(n-1)(1+k/2)
\quad\text{and}\quad
n(1+k/2)+\Theta(k^2)
\]
under their convention, and they show that even critical \(k\)-trusses still admit sparse constructions [1806.05523]. A plausible implication is that any refined \((k,P)\)-truss should be evaluated not only by its local support semantics but also by what global structures it can or cannot exclude.

Taken together, these works support a precise but non-unified understanding of the refined \((k,P)\)-Truss Model. It is best viewed as a generalized truss framework in which \(k\) remains the cohesion threshold, while \(P\) specifies the support context: personalized visibility, hop-bounded common-neighbor structure, simplex-dimensional support, or, more tentatively, resilience under support loss. The family resemblance lies in subgraph- or subcomplex-relative support, monotone decomposition, and trussness indexing; the main divergence lies in what counts as support and what object—edge, personalized edge set, or simplex—receives trussness [1906.00140; 2411.06405; 2102.07564; 1906.12335].

Source: https://www.emergentmind.com/topics/refined-k-p-truss-model