WorksComputer Science
Statistical Detectability of Coordination
Abstract
Information operations exploit coordinated activity on social media to amplify messages and manipulate public discourse. A common approach to coordination detection is to project account-entity interactions into account similarity networks based on temporal proximity, action counts, or behavioural similarity, and identify dense subgraphs or behavioural clusters. However, the current methodology finds groups of co-activated accounts or partitions all users into most similarly acting clusters, even when the accounts are acting independently of each other or the co-activity is due to some expected source. We formulate coordination detection based on hypothesis testing of account pairs and compare different null models. We connect the resulting pairwise scores to the planted-submatrix detection problem and derive statistical and computational limits as functions of group size, population size, and signal strength. We estimate the signal for 7 modalities of coordination over 26 labelled information-operation datasets, and compare the predicted regime with the performance of recovery heuristics. Our method achieves higher agreement with the labelled ground truth than previously reported, but the choice of null model is important and it changes the regime of most modalities. While it is always possible to extract a set of significantly densely co-acting accounts across nearly all modalities regardless of regime, density alone is uninformative of detection performance, rendering anomalous cluster discovery a poor proxy for recovering an information operation. Detectability boundaries establish whether an operation of a given size and signal can be recovered at all, and how much confidence a flagged group warrants.
Introduction
Information operations exploit coordinated activity on social media to manipulate conversations and spread disinformation through synchronised actions, such as simultaneous hashtag usage, coordinated replying, or orchestrated content amplification [1, 2, 3, 4]. Detecting such orchestrated campaigns from their online traces is critical to preserve the integrity of online spaces.
To isolate coordinated groups, detection pipelines typically build a coordination network, linking accounts whose actions resemble one another or co-occur close in time, such as accounts posting the same URL, reusing the same hashtag, reposting the same post, or replying in near-synchrony [3, 5, 6, 7, 8, 9]. Clustering of account behaviour or community detection methods then recover groups whose internal connections are denser than their connections to the rest of the network or whose activity patterns are similar [10, 11, 12]. Parameters are typically tuned empirically or set from prior domain knowledge, treating “similar enough,” “close enough,” or “dense enough” as a design choice. However, the choice of time window and what kind of clusters count as synchronised determine which coordination is detectable [13, 9].
A central challenge is to separate coordination from co-activity that arises by chance or from patterns expected among organic users. For example, users can act close in time by reacting to the same trending topic or hashtag, by following their own daily rhythms, or simply because they are very active. Accounting for such activity can change conclusions: Pantè et al. [ [14] show that, once organic control accounts are included, there is no statistical evidence of the inter-state coordination that earlier studies had suggested. We argue that a detected group is evidence of coordination only if expected patterns and resulting chance co-activity cannot explain its total signal. Benchmarks on labelled data show whether a detector recovers known operations, but do not quantify how surprising the observed structure is under the assumption of independent behaviour.
We treat coordination as a hypothesis-testing problem with two levels. At the level of pairs, we test each pair of accounts against a null model with prescribed patterns but otherwise independent activity, and turn its -value into a score, so that a coordinating group of accounts among becomes a block of elevated scores in an matrix. At the level of groups, we reduce coordination detection to the planted submatrix problem, for which the conditions under which such a block can be detected are known [15, 16, 17, 18, 19]. In contrast to the typical approach of graph clustering without a statistical model, which always returns a partition of the graph and where quality functions such as modularity are well-known to be prone to overfitting Guimera et al. [ [20], Peixoto [ [21], this formulation allows us to answer under what conditions coordinated behaviour can be detected. We derive a phase diagram with three detection regimes (impossible, hard, and easy), whose boundaries are functions of group size, population size, and signal strength. We also provide a theoretical basis for the recent empirical observations that multi-modality improves coordination detection [9, 22], showing that combining independent modalities of equal signal lowers every detection threshold by a factor of , so an operation that cannot be detected in any single layer can become detectable.
We measure the signal strength of 26 labelled information operations [23] and compare the predicted regime with the recovery of two greedy heuristics and a spectral method on 7 modalities and an aggregated view of each operation. Since chance synchrony has several sources, we implement four null models, and measure how much the conclusions depend on the definition of chance. The measured signal depends on the null model, because each null model attributes different parts of the observed synchrony to chance. Our approach has several assumptions that are common in the literature, but are often implicit, and made explicit by our principled statistical framework. This allows us to test the effect of these various assumptions to the peformance of the coordination detection.
This paper makes several contributions by formulating coordination detection as hypothesis testing within the planted-submatrix problem. First, we derive information-theoretic and computational boundaries for the detection of a coordinated campaign. Second, we introduce a new methodology for coordination detection, which achieves higher agreement with the labelled ground truth than previously reported [22], although the evaluation protocols cannot be made exactly equivalent. Third, we evaluate four null models on 26 information operations and show how and why the predicted regimes depend on the choice of the null model. Fourth, we show that multi-modal analysis lowers detection thresholds by . Finally, we show that a permutation test on the density of a returned set does not indicate recovery: the heuristics return significantly dense sets even when they do not recover the operation. Thus, structural properties of the coordination networks alone are an insufficient proxy for coordination detection.
Related Work
Coordination detection. Most methods for coordination detection build a network in which an edge links two accounts that act alike, and extract groups from the resulting network by connected components, community detection, or clustering [3, 5, 10, 8]. Such pipelines have been applied to elections [5, 24], to information operations [8, 11], and across platforms [25]. The modality, the size of the time window, and the threshold to retain significant edges are usually design choices [12]. The clustering and community detection methods used in this step return a partition or a group for any input network, including random graphs. Commonly used quality metrics such as modularity are known to overfit: modularity can take high values even on random graphs [20], so a high modularity is not evidence of communities [21]. The question is therefore not only whether a coordinating cluster exists, but whether its members co-act more than independent accounts would. A few studies test suspicious accounts against control accounts [14] (Mungari et al. 2026), or compare the coordination of accounts in an information operation with samples of regular users [Keller et al., 2020]. Still, the field lacks a formal and statistically grounded definition of coordination, including well-defined null models [12], and there is limited knowledge of when coordination is even detectable.
Detecting a planted submatrix. Like other methods that extract groups of nodes, dense-subgraph extraction returns a dense set for every input, including pure noise [Charikar, 2000, Hooı et al., 2016, Shabalin et al., 2009]. Statistical work asks when a planted block can be detected. Scan statistics detect it above the information-theoretic boundary [Kulldorff, 1997, Arias-Castro et al., 2011, Addario-Berry et al., 2010], while polynomial-time methods need a larger block or a stronger signal [17]. Locating the block, i.e. recovery, is at least as hard as detecting it [17]. Polynomial-time recovery is known above a signal of the order of the spectral boundary [Hajek et al., 2018], and the statistical and computational limits of recovery are also known for several planted submatrices [Chen and Xu, 2016]. The computational lower bound holds for broad classes of distributions of the entries of the matrix [18]. To our knowledge, these results have not yet been applied to coordination detection.
Methods
Co-action layers and null models
Figure 1. The -value of a pair under the account-activity null model. (a) Daily activity profiles of two accounts. (b) Two users co-act. (c, e) Two patterns of co-action times. (d, f) Distribution of the time intervals between co-actions, observed and under the null model, with the -value; the dotted line at 3 minutes shows that counts over a fixed time window might not distinguish the two co-action patterns.
We model the activity of accounts. An item is the object of an action, and a modality is a type of action on an item. The available modalities depend on the social media platform. In Section 4 we use seven modalities: using a hashtag, sharing a URL, mentioning an account, replying to a post, replying to an account, reposting a post, and reposting an account. Two accounts co-act in a modality if both act on the same item. Each modality defines a layer, a graph whose edges connect accounts that co-act at least once. Such pairs of accounts are tested for temporal synchronisation, and each edge carries a -value that measures how unusually the pair acts on their shared items, under a null model.
Accounts that coordinate have been observed repeatedly performing similar or identical actions within a short span of each other, more often than other accounts [3, 5, 8]. Making this precise requires a choice of how short “short” is, yet different modalities operate on different natural time scales [22], and there is no consensus on a single window value that works across modalities or across documented coordinated campaigns [13, 12].
For a tested pair and a shared item , let and denote their respective action timestamps on . For each , the interval to the next action of is
provided such an action exists. We define as the mean of the existing intervals in both directions. This statistic requires no fixed time window.
A null model generates randomised replicates of the actions on . With item-level Monte Carlo replicates, , where counts replicates whose mean interval does not exceed the observed . We use Fisher’s method to combine the -values of all items shared by the pair in modality into .
We consider four null models that preserve each account’s actions on each item but randomize their timestamps to test for synchrony. We can approximately order the null models by the amount of observed temporal structure that they preserve [Gavin et al., 2022]. The uniform-modality null model draws the time of each action uniformly from an evenly spaced grid with one point per action, between the first and the last action in the modality. The uniform-account null model instead uses a separate grid per account, spanning its first and last actions across all modalities, thus preserving each account’s active period. Both null models attribute every other timing pattern, including daily rhythms and reactions to the same item, to synchrony. The account-activity null model preserves the date of each action and redraws its hour from the hourly profile of the account for that day of the week. Two accounts being active at the same hours is then not by itself evidence of coordination, if it is compatible with their baseline activity profiles. This null model does not model reactions to an item, so it overstates the synchrony of accounts that respond to the same post or URL soon after it appears (Section 4). The item-virality null model redraws the action times of an account on item uniformly, with replacement, from the action times of all accounts on . This keeps the adoption curve of each item in expectation, so accounts that react to a viral item at the same time are not by themselves evidence of coordination.
Coordination as planted submatrix
We model coordination as a single group of accounts whose activity produces pairwise independent -values from a coordination distribution , while all other pairs produce -values from a uniform distribution on . This model defines a hypothesis, , which is tested against the null (absence of coordination). Our modeling makes the following two assumptions. First, the -values are independent, treating all the dependencies in the activity signals of nodes as encoded in the null model. Second, we assume that there is no internal structure among the coordinated or uncoordinated nodes. The latter is analogous to the typical, but often implicit, assumption in community detection methods that the clusters are internally homogeneous Fortunato et al. [2010].
For a coordinating pair of accounts we assume with . This is a common family of generating -values under an alternative hypothesis because its non-increasing density assigns greater probability to small -values [Sellke et al., 2001]. Moreover, Fisher’s method is optimal for combining independent -values following this family [Heard and Rubin-Delanchy, 2018]. However, our analysis can be carried out for any other continuous distribution supported on . The log-likelihood ratio of against the uniform distribution, evaluated at the observed , is , which is increasing in . For a fixed set of independent pairs sharing the same , rejecting when the sum of exceeds a critical value is uniformly most powerful: among tests with the same significance level, it has the highest probability of detecting coordination for every [Lehmann and Romano, 2005].
This motivates the centred score
which has mean 0 and variance 1 under . Under the Beta alternative, , with mean , which we call the signal strength. The scores depend on the choice of null model (Section 4) and define, for each layer , a symmetric matrix where the diagonal is left unspecified.
Let be the family of all subsets of accounts. Under , the coordinating group is drawn uniformly from . Assuming independence across pairs under and conditional independence given under , the off-diagonal entries follow for :
Our formulation is a variant of the well-studied planted-submatrix problem, for which regimes of statistical impossibility, computational hardness, and polynomial-time detection have been characterised asymptotically. To derive the corresponding detection boundaries for our setting, we must account for two differences from existing formulations. First, our matrix is symmetric, with independent entries and independent entries in the planted block. The formulations of Butucea and Ingster [2013] and Ma and Wu [2015] allow different sets of planted rows and columns, while Brennan et al. [2019] consider a single set, as we do, but treat all entries as independent. In contrast, our score matrix is symmetric, with an unspecified diagonal, and therefore has only independent entries, of which belong to the planted block.
The computational lower bound of Brennan et al. [2019] transfers to our problem by a reduction. Given an instance of their problem, we construct by setting for . This polynomial-time reduction preserves the distribution under both hypotheses. Thus, a polynomial-time test for our problem would
also solve theirs, contradicting the lower bound under the planted clique hypothesis. The reduction transfers the lower bound to our problem, but it does not establish its tightness: the reduction discards entries of their block on and below the diagonal. We obtain computational achievability thresholds from the sum and spectral tests, and derive the information-theoretic boundary for our symmetric model in Section 3.3. Second, unlike the classical Gaussian formulations, our entries follow shifted exponential distributions. We derive the upper bounds from the moment generating function of and the information-theoretic lower bound from the second moment of the likelihood ratio (Appendix A).
Detection boundaries
Next we derive sufficient conditions for detecting the presence of a planted block, using exhaustive search and polynomial-time methods, as functions of the signal strength , the total number of accounts , and the block size .
Exhaustive search. The test searches over all possible blocks of size and rejects when
exceeds a critical value , chosen so that the probability of a false detection as . We obtain the critical value from a Chernoff bound and a union bound over as for any fixed (Appendix A.2). Under , the true coordinating group has expected total score , so the scan detects the block for when , as for Gaussian entries [15].
Information-theoretic boundary. The total error probability of a test is the sum of the probabilities of a false positive under and of a missed detection under . Below the information-theoretic boundary, no test can tell from with total error probability below a fixed positive level, regardless of the computational cost [15, 16, 18]. Following the second-moment method [15] (Ingster and Suslina 2012; Addario-Berry et al. 2010; Arias-Castro et al. 2011; Arias-Castro and Verzelen 2014), we bound the total error probability from below through the second moment of the likelihood ratio of against : if for a constant , every test has a total error probability of at least [Tsybakov and Zaiats, 2009]. The boundary therefore depends on the error level that one requires, through . We take because with this value the information-theoretic boundary for large groups equals the sum boundary derived below. With this choice, every test has a total error probability of at least below the signal at which , which is, to leading order (Appendix A.4),
Another changes the large-group term of Equation (2) to , while the small-group term does not depend on (Appendix A.4). For small groups, is a factor below the scan boundary ; for between the two, it is not known whether any test detects the block (Appendix A.4).
Computational boundary. The computational boundary is the signal strength above which a polynomial-time test detects the block and below which no polynomial-time test does [16, 18] (Hajek et al. 2015). As in [16] and [18], we bound it from above with polynomial-time tests. The sum test rejects when the sum of all scores is large [15] (Arias-Castro and Verzelen 2014). Under , has mean and standard deviation ; under , its mean is , and near the boundary its standard deviation remains close to (Appendix A.5). As in [15], we divide by its standard deviation under . The normalised sum has expected value under and under . For Gaussian entries, the total error of the sum test tends to when , and the boundary lies where is approximately one [15]. Following [15], the sum boundary is the signal strength at which (Appendix A.5):
The spectral test uses the largest eigenvalue of the score matrix. Under , the block adds to a matrix with entry on the pairs of the block, whose largest eigenvalue is ; if this eigenvalue is large enough, it raises the largest eigenvalue of above its value under . As in [Montanari et al., 2015], the test rejects when the largest eigenvalue of exceeds with , and it detects the block if and only if [Montanari et al., 2015, Knowles and Yin, 2013] (Appendix A.6).
Since each test detects the block above its own boundary, a polynomial-time test boundary is
The sum test gives the minimum when , and the spectral test otherwise. For , no polynomial-time test detects the block under the planted clique hypothesis when [18], where is the Kullback–Leibler divergence of the distribution of a block entry from the distribution of a null entry. For our scores, , so the condition is (Appendix A.7). For , the lower bound of [18] applies to our scores only for of order at most , far below (Appendix A.7). For between these two values, the bounds do not determine whether a polynomial-time test detects the block.
Regimes. Detection is impossible if , hard if , and easy if . For , hard thus means that neither the sum test nor the spectral test detects the block. With and , the three boundaries are, to leading order, the lines (scan and small-group information-theoretic boundary), (spectral), and (sum and large-group information-theoretic boundary). Up to terms that vanish for , these are the coordinates of Brennan et al. [2019]. The lines and meet at : for larger groups the information-theoretic boundary equals the sum boundary, so detection is easy.
Untested pairs. When two accounts do not co-act, their pair has no -value. We give such pairs the score under both hypotheses, so the signal strength can still be computed as the mean score over all pairs of the group, tested or not. The boundaries above assume that every pair is tested. With untested entries, the sum of all scores keeps its mean under and has a smaller variance under both hypotheses, so the sum test detects the block at a smaller than ; the same holds for the spectral test when the untested pairs are spread uniformly. The sum and spectral boundaries of a matrix with untested entries are therefore at most the values for a complete matrix. A layer in which is easy, but a layer in which is below or may still lie above the corresponding boundary of its matrix with untested entries, so its regime, impossible or hard, is an approximation.
Other distributions for the alternative hypothesis . The assumption can be relaxed to any continuous distribution on . The scan, sum, and spectral tests detect the block when exceeds their boundaries, for any continuous under which the variance of stays bounded. The scan rejects when the largest sum of scores over a set of accounts exceeds its critical value, and the sum test when the sum of all scores exceeds its critical value. Both critical values are computed from the null distribution of the scores, so they do not depend on . Under , the sum over the block has mean and variance , and the central limit theorem these two numbers determine its distribution when is large. The spectral test uses the largest eigenvalue of the score matrix , which we can write as . The expected matrix has for every pair in the block and elsewhere, and its largest eigenvalue is . The entries of outside the block have mean and variance , and the largest eigenvalue of such a matrix is close to [Bai and Yin, 1988]. When , the largest eigenvalue of is close to , which exceeds the edge of the spectrum [Féral and Péché, 2007]. The distribution only affects the block entries of . If has a finite fourth moment, the derivation is similar to Appendix § 6.6, and, when , the largest eigenvalue of has the same limits as in Appendix § 6.6.
The information-theoretic boundary follows from the second moment of the likelihood ratio under independence of -values, and depends on through , where is the density of . The computational boundary assumes the planted clique hypothesis and requires conditions on , so it depends on through the Kullback–Leibler divergence. Two distributions with the same can have different , and gives only the bound . Hence a layer with is easy for any . A layer with below the information-theoretic boundary or below is impossible or hard only if and are close to ; otherwise a test that uses can detect the block.
Several modalities. A group that coordinates in modalities forms a block in each of layers. Assume the simplest case where the layers carry the same signal , their scores are independent, and every pair is tested in all layers. Then, the summed score has null mean and null variance , and its block entries have mean . By dividing by , the summed score has null variance and the mean of the block becomes . This division does not change the decision of any test. The summed matrix therefore behaves as one layer with signal , and every boundary is multiplied by : a coordinating group whose signal is below the information-theoretic boundary in each modality can be detectable when the signal is aggregated over many modalities. The null distribution of the summed score is a shifted Gamma distribution with shape parameter , so the factor is exact for the sum and spectral boundaries and holds to leading order for the scan boundary and the information-theoretic boundary. A simulation of the scan with , , and is consistent with this result: the signal at which the scan has power at level has a fitted slope against on a log–log scale between and for , closest to the limit for the largest .
Signal Estimation and Group Detection
The layer graph of an operation and modality under one null model has as nodes the accounts that co-act with at least one other account. Let be the number of labelled accounts of the operation, and be the set of labelled accounts in the layer; , since a member that does not co-act in the modality is not in the layer. We estimate the signal strength of the block by the mean score of the labelled pairs of the layer, treating as the planted set ,
where an untested pair contributes . A layer with has no measurable signal, because its labelled pairs score no higher on average than independent pairs; we report it as no signal. The control mean of a layer is computed like over the pairs of control accounts, and is under a correctly specified null model.
We cap the scores at , because the far tail of an estimated null model is not validated and a few extreme pairs would otherwise determine . The cap is above the largest score among null pairs: the largest layer has tested pairs, and the expected number of null pairs above the cap is . One shared item gives a score of at most , so a pair reaches the cap only if it shares at least nine items. The cap affects between to of the tested pairs of a layer.
Recovery heuristics. The scan over all sets is computationally infeasible for most data sets, so we define two greedy heuristics. Size-agnostic peeling repeatedly removes the node whose removal most increases the mean score of the remaining set, and returns the set of at least accounts that maximises the density . The fixed-size heuristic is given the size of the block in the layer, seeds a set of nodes from the peeling order, and improves by single swaps (at most 200 iterations), keeping the best of 3 restarts. A permutation test compares the density of the returned set with the densities of the sets that the same heuristic returns after the edge weights are shuffled (100 replicates). The returned set beats chance if a random set of nodes contains at least as many labelled nodes with probability below (hypergeometric test). We also report the score of against the labelled nodes, and the signal of the returned set, computed as in Equation (3) with in place of . For the spectral method, we compute the eigenvector of the largest eigenvalue of the score matrix of the layer, choose its sign as the sign whose largest entries have the larger total score, and return the accounts with the largest entries. A permutation test compares with its values after the edge weights are shuffled (100 replicates).
Results
The predicted regime of the 26 labelled information operations depends strongly on the chosen null model (Figure 2, Table 2). The item-virality null model places the fewest layers in the easy regime (36%), followed by the account-activity null model (54%), the uniform-account null model (81%) and the uniform-modality null model (84%; Table 1). These differences stem from the fact that the null models differ in which timing patterns they attribute to chance. The item-virality null model redraws the action times from the adoption curve of the item. Because the data contain only a sample of control accounts, many of the post and URL items have exactly two adopters, and on such an item the item-virality null model gives when each account acts once. As a result, this null model measures no signal in 50% of layers, and the -values of the labelled pairs follow Beta(, 1) with a median of 0.98, close to the uniform distribution. The account-activity null model keeps the hourly profile of each account, but it does not reproduce the independent reactions of different accounts to a new popular item, so the control mean is slightly positive in all 71 reply (post), repost (post), and URL layers. The uniform null models keep only the period in which a modality or an account is active. They measure the strongest signal for the information operations (see Table 2 and Figure 2b). However, their control mean is positive in almost all layers, and the -values of the control pairs are further from uniform than under the account-activity null model (see Figure 2c, median of 0.37 and 0.36, against 0.74). That is, the choice of a null model is a compromise between distinguishing the coordinated group and the control group: a null model that attributes more timing patterns to chance follows better the uniform -value distribution for control accounts, but also absorbs the coordination itself, while one that attributes fewer patterns leaves organic synchrony in the scores but separates the coordinated pairs more strongly.
Figure 2. (a) Phase diagrams under the four null models (uniform-modality, uniform-account, account-activity, item-virality). Each point is a graph layer, coloured by the of the set returned by the spectral method. (b, c) Beta distributions fitted by maximum likelihood to the -values under each null model. (b) Density of estimates for Beta(, 1) fitted to the pairs labelled as coordinated. (c) Density of estimates of Beta() fitted to -values between control pairs or mixed pairs, as contours at 5%, 35%, and 70% of its maximum; the + marker shows the uniform distribution, expected under a calibrated null model.
| regime | layers | beats chance agn. | beats chance fix. | beats chance spec. | median agn. | median fix. | median spec. | test rejects agn. | test rejects spec. |
| uniform-modality null model | |||||||||
| impossible | 6 | 0 | 0 | 0 | 0.00 | 0.00 | 0.00 | 6 | 6 |
| hard | 25 | 4 | 6 | 10 | 0.00 | 0.00 | 0.03 | 22 | 24 |
| easy | 168 | 123 | 126 | 123 | 0.34 | 0.59 | 0.62 | 153 | 156 |
| no signal | 0 | – | – | – | – | – | – | – | – |
| uniform-account null model | |||||||||
| impossible | 10 | 1 | 3 | 0 | 0.00 | 0.00 | 0.00 | 10 | 10 |
| hard | 26 | 10 | 11 | 12 | 0.01 | 0.03 | 0.02 | 24 | 25 |
| easy | 162 | 132 | 139 | 134 | 0.35 | 0.67 | 0.66 | 157 | 156 |
| no signal | 1 | 0 | 0 | 0 | 0.01 | 0.01 | 0.00 | 1 | 1 |
| account-activity null model | |||||||||
| impossible | 51 | 15 | 15 | 18 | 0.00 | 0.00 | 0.00 | 47 | 46 |
| hard | 24 | 8 | 7 | 7 | 0.01 | 0.00 | 0.00 | 21 | 21 |
| easy | 107 | 82 | 89 | 76 | 0.26 | 0.59 | 0.56 | 105 | 105 |
| no signal | 17 | 7 | 10 | 10 | 0.08 | 0.22 | 0.17 | 17 | 16 |
| item-virality null model | |||||||||
| impossible | 14 | 3 | 2 | 4 | 0.00 | 0.00 | 0.00 | 14 | 14 |
| hard | 13 | 6 | 4 | 8 | 0.01 | 0.00 | 0.04 | 13 | 13 |
| easy | 72 | 42 | 38 | 39 | 0.17 | 0.28 | 0.35 | 72 | 72 |
| no signal | 100 | 33 | 27 | 40 | 0.03 | 0.12 | 0.23 | 85 | 87 |
Table 1. Recovery by predicted regime: layers in which the returned set beats chance, median against the labelled accounts in the layer, and layers in which the permutation test rejects (; for the spectral method, the test on ). agn.: size-agnostic; fix.: fixed-size; spec.: spectral.
| layer | control mean , median [range] account | control mean , median [range] item | control mean , median [range] unif. | control mean , median [range] modality | coordination , mean [range] account | coordination , mean [range] item | coordination , mean [range] unif. | coordination , mean [range] account |
| pooled | 0.012 [0.005, 0.076] | 0.006 [, 0.16] | 0.215 [0.061, 1.48] | 0.114 [0.040, 0.56] | 1.13 [0.046, 3.56] | 1.10 [, 5.67] | 5.78 [0.93, 15.09] | 4.04 [0.48, 11.78] |
| mentions | 0.001 [, 0.014] | [, 0.12] | 0.098 [0.010, 1.27] | 0.032 [0.003, 0.44] | 0.56 [, 7.34] | 0.54 [, 4.09] | 4.04 [0.53, 15.03] | 2.71 [0.25, 14.75] |
| hashtags | 0.011 [0.004, 0.037] | 0.001 [, 0.17] | 0.167 [0.041, 0.50] | 0.075 [0.018, 0.27] | 0.65 [, 3.08] | 1.23 [, 7.57] | 4.79 [0.52, 14.38] | 3.34 [, 11.10] |
| reply (acct) | 0.000 [, 0.004] | [, ] | 0.008 [0.003, 0.089] | 0.002 [, 0.025] | 0.062 [, 0.55] | [, 0.56] | 1.64 [0.22, 6.02] | 0.99 [0.091, 2.95] |
| reply (post) | 0.016 [0.004, 0.28] | [, ] | 0.058 [0.012, 0.52] | 0.024 [0.006, 0.41] | 0.44 [0.009, 2.30] | [, ] | 1.17 [0.019, 3.80] | 1.07 [0.019, 3.72] |
| repost (acct) | 0.001 [, 0.014] | [, 0.046] | 0.052 [0.006, 0.76] | 0.018 [0.002, 0.25] | 0.48 [, 6.56] | 0.29 [, 2.95] | 3.43 [0.17, 13.83] | 2.41 [0.091, 12.43] |
| repost (post) | 0.005 [0.0005, 0.062] | [, ] | 0.015 [0.002, 0.18] | 0.009 [0.0009, 0.12] | 1.61 [0.043, 7.38] | [, 0.19] | 3.55 [0.16, 14.61] | 3.43 [0.15, 14.60] |
| URLs | 0.005 [0.0004, 0.13] | [, ] | 0.021 [0.002, 0.52] | 0.010 [0.0009, 0.23] | 0.85 [0.032, 3.63] | [, 0.47] | 2.43 [0.18, 9.77] | 2.17 [0.16, 9.49] |
| all layers | 0.005 [, 0.28] | [, 0.17] | 0.055 [0.002, 1.48] | 0.021 [, 0.56] | 0.73 [, 7.38] | 0.37 [, 7.57] | 3.43 [0.019, 15.09] | 2.57 [, 14.75] |
Table 2. Control mean (median and range over the operations) and coordination of the labelled pairs (mean and range over the operations) by modality under each null model. The control mean is 0 under a correctly specified null model.
Recovery of the ground-truth groups using the spectral method exceeds that of all 20 methods that Iannucci et al. [ [22] evaluate on the same 26 operations (Figure 3). As in their evaluation, we compute against all labelled accounts of the operation. Both evaluations use the ground truth beyond the final score: the spectral method is given the size of the labelled group in the layer, and [22] is the of the cluster that best matches the labelled accounts. When each layer and null model is considered separately, the spectral method exceeds all 20 methods in 10 of the 28 single-modality combinations. The highest median is 0.62 for reposts of a post under the uniform-account null model, compared to 0.39 for synchronised hashtags [10] and 0.34 for time-aware (RT) [22], which uses the same modality. Some of the layers carry poor signal over the operations, but even when evaluated over all operations and single-modality layers, the spectral method reaches a median of 0.40 under the uniform-account null model, against 0.36, 0.24, and 0.18 under the uniform-modality, account-activity, and item-virality null models, and against 0.39, the highest reported median [Iannucci et al., 2022], achieved by synchronised hashtags. On the layers that the theory places in the easy regime, the spectral method exceeds all of the previous methods [22] under every null model, with a median of 0.45, 0.48, 0.46, and 0.39 under the uniform-modality, uniform-account, account-activity, and item-virality null models, and a median recall of 0.34 to 0.38; on the hard and impossible layers, the median is at most 0.04.
Figure 3. of the spectral method for each null model and layer that scores higher than the top 3 methods [22].
Under all four null models, combining modalities moves operations toward the easy regime, but with less than the factor derived in § 3.3. We combine the modalities in two ways. The summed layer adds the scores of a pair over the modalities, and is placed on the phase diagram at . The pooled layer combines the -values of a pair across all modalities by Fisher’s method, so, under our assumptions, every tested pair has expectation 0 and variance 1 under , independently of the number of layers. Since Fisher’s statistic equals, up to a constant, the sum of the scores, the two layers differ only in how the sum is standardised. Under the account-activity, item-virality, uniform-modality, and uniform-account null models, the pooled layer is easy in 18, 18, 23, and 22 of the 26 operations, and the summed layer in 17, 14, 22, and 22, in both cases more often than the single-modality layers. The regime of the pooled layer also depends less on the null model: the four null models agree on 65% of operations, against 22% of the single-modality layers. However, the boundaries move less than the factor for two reasons. First, our derivation requires every pair to be tested in all layers, while a pair is tested in a mean of 1.34 layers (median over the
operations; range 1.04 to 2.0). Second, our derivation assumes one block with the same signal in every layer, while, in practice, we observe that each modality involves a different subset of the operation. Thus, a combined layer averages the signal of the strongest modality with that of the others. Under no null model is an operation easy on the pooled layer without being easy in some single modality. Recovery is lower on the combined layers: the median of the fixed-size heuristic is 0.03, 0.19, 0.12, and 0.02 on the pooled layer and 0.18, 0.39, 0.29, and 0.02 on the summed layer, against 0.60, 0.69, 0.64, and 0.37 on the easy single-modality layers (uniform-modality, uniform-account, account-activity, and item-virality null models), although a combined layer is scored against nearly the whole operation in a larger population.
Under all four null models, the heuristics find a set denser than chance in almost every layer, whether or not this set is the operation (Table 1). Even on layers where the found set does not beat chance recovery, the found set is still denser than chance in the vast majority of cases. The sets found by the size-agnostic heuristic, fixed-size heuristic, and spectral method are also denser than the operation in most layers. The pooled layer falls under the easy regime in most operations under every null model, but the spectral set has a median precision of 0.06, 0.22, 0.10, and 0.04, and a signal 1.9, 1.5, 1.5, and 3.8 times that of the labelled pairs. The scores therefore contain dense groups of mostly control accounts, which the model of one block in independent scores excludes. A significant permutation test only shows that a layer contains a dense set, but does not guarantee that the dense set represents the information operation.
Conclusion
We treat coordination detection as hypothesis testing across all accounts, reducing the task to detecting a planted block within a matrix of scores. Our derived boundaries establish the signal thresholds below which detection, and therefore recovery, is impossible, and above which polynomial-time tests detect the block. Our derivation shows how combining scores across multiple modalities lowers the statistical and computational boundaries. We demonstrate that algorithmic recovery is bounded by the chosen null model, through the timing patterns that it attributes to chance and the signal that it can measure. Across 26 labelled information operations, null models that attribute fewer patterns to chance are less calibrated, but they separate coordinated from organic pairs more strongly. Consequently, under these models, a spectral method recovers the operations better, by median , than previously proposed coordination detection methods (Figure 3).
The dependence on the null model reveals two criticalities that affect the current study on coordination detection. First, permutation tests find a significantly dense set across nearly all layers, so the existence of a dense cluster in a coordination network is not evidence of coordination. Second, the simplest null models, that keep only the active period and attribute every timing pattern to synchrony, recover the coordinated groups best. Such null models separate the labelled accounts from the control accounts only based on their timing, which points to a sampling artefact in benchmark datasets: the accounts of these operations are not hiding in the timing patterns of organic accounts. This can reflect detection bias of the platform itself, so that information operations accounts that hide better are absent from the ground truth, or the collection of the data. The control accounts were sampled by the hashtags and dates of each operation, with at most 100 posts on those dates [23], and the uniform-account null model spreads the actions of an account over its own active period, so the same co-action is far more extreme for two accounts of the operation than for two control accounts. This calls for additional benchmark datasets, so that coordination detection frameworks can be evaluated against realistic conditions and more varied strategies and patterns of coordination.
Limitations. Our approach focuses on temporal synchronisation conditional on accounts interacting with the same items, rather than coordination expressed through unusual co-adoption. The null models account separately for individual activity patterns, item virality, or only the active period, but none controls for both activity patterns and item virality simultaneously. A joint null model could address this limitation, although it would require additional modelling assumptions and risk absorbing genuine coordination into the estimated baseline. Conversely, considering each null model separately allows us to examine how different sources of chance synchronisation affect the conclusions. Exploring joint null models through stratified or weighted randomisation is a direction for future work.
Our theoretical formulation and empirical evaluation rely on simplifying assumptions. First, on a statistical level, we assume independent -values, whereas pairs that share an account are correlated. Furthermore, because Monte Carlo -values are discrete and only approximately uniform under the null model, they yield slight biases and cause score means and variances to be 0 and 1 only approximately. However, because real-world detection lacks labels, the deviation from calibration cannot simply be solved by empirical null distributions. Instead, this calls for further research on null models that accurately capture organic timing without absorbing the coordination signal itself. Second, our derivation relies on the planted submatrix problem, assuming a single, homogenous block of coordinated accounts within a fully tested matrix. Because real-world information operations may be complex and consist of multiple sub-groups, future research could extend the derivation of phase boundaries to nested block structures. Similarly, the multi-modal gain assumes a uniform group producing equal signal in all modalities. Deriving the signal gain for coordination that alternates across modalities, or modalities with unequal signal, is left for future work. Finally, since the greedy heuristics and the spectral method are empirical approximations of the exhaustive scan and polynomial-time tests used in the theoretical derivation, their failure in the easy regime reflects either algorithmic limitations or, where , the gap between detection and recovery (§ 3.3). Since the densest subgraph problem with negative edge weights is NP-hard [Tsourakakis et al., 2019], and our scores take negative values, good heuristics for weighted signed multiplex networks are essential to bridge theoretical detectability and empirical detection.
Derivation of the boundaries
Assumptions
The derivations use the model of § 3.2 for one layer, under these assumptions:
(A1) the scores , , are independent;
(A2) under , every has the uniform distribution ;
(A3) under , is uniform on with for , and otherwise;
(A4) is known; (A5) every pair is tested. In the data, a pair that does not co-act has score 0 under both hypotheses. The diagonal of is 0 for the spectral test.
, and , denote the distributions of and the expectations under , . Asymptotic statements refer to with and for fixed and ; a formula holds to leading order if the ratio of its two sides tends to 1.
Distribution of the scores
The log-likelihood ratio of against the uniform distribution, evaluated at the observed , is
The likelihood ratio of a block entry against a null entry is . As a function of the score, , the exponential of (4).
Since the score is a one-to-one function of the -value, takes the same value whether it is computed from or from . Every divergence between and and every condition in [18] depends on the two distributions only through , and therefore has the same value for the -values and for the scores.
For a set with pairs and :
with mean 0 and variance under , and mean and variance under .
Scan boundary
Following the scan-statistic analysis of Arias-Castro et al. [2011], Butucea and Ingster [ [15], we use a Chernoff bound and a union bound to derive the sufficient detection condition for the exhaustive scan stated in Section 3.3, adapting the calculation to symmetric matrices with shifted-exponential entries.
A set contains pairs, and there are such sets. For a set , let , so that , and let . As with , , and bounded , we have and .
Under , , so for the logarithm of the moment generating function of a score is
Since for ,
For a fixed set , and , Markov’s inequality applied to and the independence of the scores of give . The scan exceeds only if at least one set does, so a union bound over and give
with .
Let
and take for a fixed , and , which tends to 0 because . By (6), , and with and ,
Inserted into eq:7, this gives
As , the bracket tends to and , so .
Under , the scores of are independent with , , so and . The maximum over all sets is at least the sum over the block, . If , the event implies , and Chebyshev’s inequality gives
The condition reads , with
Since is arbitrary, for every fixed the scan with critical value has both error probabilities tending to 0 when . This is a sufficient condition: the union bound treats overlapping sets as disjoint, so the exact boundary of the scan may be lower.
Information-theoretic boundary
Following the second-moment approach of [15] (Arias-Castro and Verzelen 2014) and the testing inequalities of [Tsybakov and Zaiats, 2009], we derive the lower bound on total testing error underlying the information-theoretic bound in § 3.3 and examine its dependence on group size and signal strength.
Error bound. A test decides from the score matrix whether to reject . Let be the set of score matrices for which it rejects. The test makes a false positive with probability and misses the block with probability .
Under , the coordinating group is drawn uniformly from , so is the average, over all , of the distribution of when . The likelihood ratio of against satisfies . For every test, that is, for every set ,
The first inequality holds because for every set . The second holds because by the Cauchy–Schwarz inequality, and [Tsybakov and Zaiats, 2009]. Since averages over , the bound holds for the error averaged over the possible groups, and therefore also for the largest error over all groups, which is at least the average.
Let be the likelihood ratio of a single score of the block against a null score; its logarithm is Equation (4) (Appendix § 6.§ 2). If the group were known, the likelihood ratio of the whole matrix would be , since the entries outside have the same distribution under both hypotheses. Under the group is uniform on , so is the average of these products:
Second moment. Squaring gives a double sum over two groups and , which we can read as two independent groups drawn uniformly from :
Under the scores are independent, so the expectation is a product over pairs. A pair in neither group contributes , and a pair in exactly one group contributes . A pair in both groups has both accounts in and contributes
with . The integral is over the -value , which is uniform on under . This gives because the score is a one-to-one function of the -value, so has the same value whether it is computed from or from (Appendix § 6.§ 2). The integral is finite only for , that is, for . If the two groups share accounts, exactly pairs are in both groups. For two independent uniform groups, has the hypergeometric distribution . Hence
Every term increases with . At the sum equals , and as it tends to infinity, because of the terms with . So, for every , there is exactly one at which . We define by . For , the bound above gives a total error probability of at least for every test; this holds at finite when is computed from Equation (10).
Closed forms. For small groups () the overlap has mean . The terms and of Equation (10) have the factor and sum to ; the terms with small vanish, and the term that remains is . This term equals , with , because the two groups coincide () with probability and then share all pairs. It is of order when . As , , and because . The condition becomes , and with ,
The value of changes only by , so it does not affect the leading term of .
For large groups () the overlap has mean and standard deviation at most , so and we can replace by . The expectation in Equation (10) is , with . As , , and as , ; together with , this gives . So is approximately , which equals at . The approximation holds at this since differs from by about , replacing by changes by about , which tends to . For the value is , whose ratio to tends to .
exceeds as soon as either the term or the bulk of the overlap distribution does, so to leading order
This minimum is an approximation of the exact value of , the numerical solution of with (10), and the ratio of the two tends to 1 as . On the 199 layers of Section 4, the exact value is between 0.54 and 1.10 times the closed form (median 0.99); the largest differences occur for groups with fewer than 10 accounts.
Gap to the scan. Where equals its first term, that is, for below about , is a factor below (Appendix A.3). The smallest signal at which some test detects the block may lie above , because rare large overlaps can inflate . For Gaussian entries, a truncated second moment that removes these overlaps closes the gap to the scan [15]; this has not been done for exponential entries. For it is therefore not proved that any test detects the block, nor that none does.
Sum boundary
Our sum test is the linear test of [15], applied to the scores above the diagonal. As in [15], we divide by its standard deviation under . The scores are independent by assumption, so with , Equation (5) applied to the block and to the remaining pairs gives
The normalised sum has mean 0 and variance 1 under , and mean and variance under .
By Chebyshev’s inequality, the test has type I error at most and, for , type II error at most . For , both bounds tend to 0 when , as for Gaussian entries [15].
In [15], the boundary of the linear test is where the normalised mean shift is approximately one. Following this, is the signal strength at which :
This value also equals the large-group term of with (Appendix A.4).
For , the order of is also the computational boundary of [18], which is , up to factors that grow slower than any power of . With (Equation (12)), this is as . Below this order no polynomial-time test detects the block under the planted clique hypothesis, and above it a polynomial-time test does (Appendix A.7).
Spectral boundary
Following the spectral-edge and matrix-deformation results of Bai and Yin [1988], Knowles and Yin [2013], we derive the spectral detection threshold stated in Section 3.3 and examine the effect of the larger variance of entries within the planted block.
The spectral boundary follows from the largest eigenvalue of a random matrix with a low-rank perturbation [Bai and Yin, 1988, Knowles and Yin, 2013]. Write . Under , has the entry on the pairs of and 0 elsewhere, so its largest eigenvalue is . Let . Outside the block, the entries of are independent, with distribution , mean 0 and variance 1 (Appendix A.2). Under , the largest eigenvalue of tends to 2 [Bai and Yin, 1988]. Under , it tends to if and to 2 if [Knowles and Yin, 2013]. The test that rejects when the largest eigenvalue of exceeds therefore detects the block if , and for no if , which gives .
The block entries of have variance instead of 1, because a centred block score has the distribution of times a centred score under (Appendix A.2). So has the distribution of a matrix of centred scores under plus times the same matrix with all entries outside the pairs of set to 0. The nonzero part of this second matrix is a matrix of centred scores under , so its eigenvalues lie between about and [Bai and Yin, 1988]. By Weyl’s inequality, adding a matrix changes each eigenvalue by at most the largest absolute eigenvalue of the added matrix [Fulton et al., 2000]. The largest eigenvalue of therefore changes by at most about , which tends to 0 as , and the limits above also hold with the larger variance of the block entries.
Computational boundary
We verify the conditions under which the reduction from the planted clique problem of Brennan et al. [2019] applies to our scores. Under , with (Appendix A.2), so
and decays with the exponent , where .
Besides the condition on , the lower bound of [18] needs two more. First, must be a computable pair: can be sampled and evaluated in polynomial time, , and both KL divergences are finite. This holds because and are exponential with the same support. Second, large values of must be rare under , because the reduction from the planted clique problem in [18] is accurate only on entries whose stays below a threshold . This is measured with the Chernoff exponent , which gives for [18]. For our scores, with .
For , the second condition is that some threshold satisfies , , and . If , that is, , the threshold with and satisfies all three: ; ; and , so . For , the second condition is , with the edge density of the planted clique hypothesis. This needs , that is, .
Coordinates of the phase diagram
The phase diagram for submatrix detection in Brennan et al. [ [18] has two coordinates: the size of the block, , and the signal, , both up to factors that grow slower than any power of . In these coordinates, detection is impossible for , possible in polynomial time for , and, under the planted clique hypothesis, hard in between.
We use the same coordinates, with the signal in place of , since (11). The exponent of differs from by for , which tends to for . Each boundary has the form , with , a constant , a factor equal to or , and exponents and . With and , the boundary is . The last term is and tends to , so as the boundary becomes the line . The scan boundary and the small-group term of have , which gives ; the sum boundary and the large-group term of have , which gives , as in [18]. The spectral boundary has , which gives . This line is not in [18], where, for , detection is possible in polynomial time already for . Layers with and are therefore hard for the sum and spectral tests.
References
- [1]Kate Starbird, Ahmer Arif, and Tom Wilson. Disinformation as collaborative work: Surfacing the participatory nature of strategic information operations. Proceedings of the ACM on Human-Computer Interaction, 3(CSCW): 1–26, 2019.DOI
- [2]Lynnette Hui Xian Ng and Kathleen M Carley. A combined synchronization index for grassroots activism on social media. arXiv:2212.13221, 2022.arxiv.org/abs/2212.13221
- [3]Diogo Pacheco, Pik-Mai Hui, Christopher Torres-Lugo, Bao Tran Truong, Alessandro Flammini, and Filippo Menczer. Uncovering coordinated networks on social media: methods and case studies. In Proceedings of the international AAAI conference on web and social media, volume 15, pages 455–466, 2021.arxiv.org/abs/2001.05658
- [5]Leonardo Nizzoli, Serena Tardelli, Marco Avvenuti, Stefano Cresci, and Maurizio Tesconi. Coordinated behavior on social media in 2019 uk general election. In Proceedings of the international AAAI conference on web and social media, volume 15, pages 443–454, 2021.arxiv.org/abs/2008.08370
- [6]Derek Weber and Frank Neumann. Amplifying influence through coordinated behaviour in social networks. Social Network Analysis and Mining, 11(1):111, 2021.DOI
- [7]Karishma Sharma, Yizhou Zhang, Emilio Ferrara, and Yan Liu. Identifying coordinated accounts on social media through hidden influence and group behaviours. In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, pages 1441–1451, 2021.arxiv.org/abs/2008.11308
- [8]Lorenzo Cima, Lorenzo Mannocci, Marco Avvenuti, Maurizio Tesconi, and Stefano Cresci. Coordinated behavior in information operations on twitter. IEEE Access, 12: 61568–61585, 2024.DOI
- [9]Lorenzo Mannocci, Stefano Cresci, Matteo Magnani, Anna Monreale, and Maurizio Tesconi. Multimodal coordinated online behavior: Trade-offs and strategies. arXiv:2507.12108, 2025.arxiv.org/abs/2507.12108
- [10]Thomas Magelinski, Lynnette Ng, and Kathleen Carley. A synchronized action framework for detection of coordination on social media. Journal of Online Trust and Safety, 1(2), 2022.
- [11]Luca Luceri, Valeria Pantè, Keith Burghardt, and Emilio Ferrara. Unmasking the web of deceit: Uncovering coordinated activity to expose information operations on twitter. In Proceedings of the ACM Web Conference 2024, pages 2530–2541, 2024.
- [12]Lorenzo Mannocci, Michele Mazza, Anna Monreale, Maurizio Tesconi, and Stefano Cresci. Detection and characterization of coordinated online behavior: A survey. ACM Computing Surveys, 58(16):1–38, 2026.arxiv.org/abs/2408.01257
- [13]Georgios Panayiotou, Lorenzo Mannocci, and Maurizio Tesconi. Setting the clock: Evaluating temporal window parameters for coordinated behavior detection. arXiv preprint arXiv:2609.21959, 2026.arxiv.org/abs/2609.21959
- [14]Valeria Pantè, David Axelrod, Alessandro Flammini, Filippo Menczer, Emilio Ferrara, and Luca Luceri. Beyond interaction patterns: Assessing claims of coordinated inter-state information operations on twitter/x. In Companion Proceedings of the ACM on Web Conference 2025, pages 1234–1238, 2025.arxiv.org/abs/2502.17344
- [15]Cristina Butucea and Yuri Ingster. Detection of a sparse submatrix of a high-dimensional noisy matrix. Bernoulli, pages 2652–2688, 2013.DOI
- [16]Zongming Ma and Yihong Wu. Computational barriers in minmax submatrix detection. The Annals of Statistics, pages 1089–1116, 2015.DOI
- [17]Tony Cai, Tengyuan Liang, and Alexander Rakhlin. Computational and statistical boundaries for submatrix localization in a large noisy matrix. 2017.DOI
- [18]Matthew Brennan, Guy Bresler, and Wasim Huleihel. Universality of computational lower bounds for submatrix detection. In Conference on Learning Theory, pages 417–468. PMLR, 2019.arxiv.org/abs/1902.06916
- [19]Matthew Brennan and Guy Bresler. Reducibility and statistical-computational gaps from secret leakage. In Conference on Learning Theory, pages 648–847. PMLR, 2020.arxiv.org/abs/2005.08099
- [20]Roger Guimera, Marta Sales-Pardo, and Luís A Nunes Amaral. Modularity from fluctuations in random graphs and complex networks. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 70(2):025101, 2004.DOI
- [21]Tiago P Peixoto. Descriptive vs. inferential community detection in networks: Pitfalls, myths and half-truths. Cambridge University Press, 2023.
- [22]Letizia Iannucci, Elisa Muratore, Antonis Matakos, and Mikko Kivelä. Detecting coordinated activities through temporal, multiplex, and collaborative analysis. In Proceedings of the International AAAI Conference on Web and Social Media, volume 20, pages 1063–1082, 2026.
- [23]Ozgur Can Seckin, Manita Pote, Alexander C Nwala, Lake Yin, Luca Luceri, Alessandro Flammini, and Filippo Menczer. Labeled datasets for research on information operations. In Proceedings of the International AAAI Conference on Web and Social Media, volume 19, pages 2567–2574, 2025.arxiv.org/abs/2411.10609
- [24]Serena Tardelli, Leonardo Nizzoli, Marco Avvenuti, Stefano Cresci, and Maurizio Tesconi. Multifaceted online coordinated behavior in the 2020 us presidential election. EPJ Data Science, 13(1):33, 2024.DOI
- [25]Federico Cinus, Marco Minici, Luca Luceri, and Emilio Ferrara. Exposing cross-platform coordinated inauthentic activity in the run-up to the 2024 us election. In Proceedings of the ACM on Web Conference 2025, pages 541–559, 2025.DOI


