arXiv subjects

Filippo Radicchi

Publications and source records attributed to Filippo Radicchi.

At least 73 records · Page 4Linked to original sources

arXiv2018

Decoding communities in networks

According to a recent information-theoretical proposal, the problem of defining and identifying communities in networks can be interpreted as a classical communication task over a noisy channel: memberships of nodes are information bits erased by the channel, edges and non-edges in the network are parity bits introduced by the encoder but degraded through the channel, and a community identification algorithm is a decoder. The interpretation is perfectly equivalent to the one at the basis of well-known statistical inference algorithms for community detection. The only difference in the interpretation is that a noisy channel replaces a stochastic network model. However, the different perspective gives the opportunity to take advantage of the rich set of tools of coding theory to generate novel insights on the problem of community detection. In this paper, we illustrate two main applications of standard coding-theoretical methods to community detection. First, we leverage a state-of-the-art decoding technique to generate a family of quasi-optimal community detection algorithms. Second and more important, we show that the Shannon's noisy-channel coding theorem can be invoked to establish a lower bound, here named as decodability bound, for the maximum amount of noise tolerable by an ideal decoder to achieve perfect detection of communities. When computed for well-established synthetic benchmarks, the decodability bound explains accurately the performance achieved by the best community detection algorithms existing on the market, telling us that only little room for their improvement is still potentially left.

physics.soc-ph↗

arXiv2017

Optimal percolation on multiplex networks

Optimal percolation is the problem of finding the minimal set of nodes such that if the members of this set are removed from a network, the network is fragmented into non-extensive disconnected clusters. The solution of the optimal percolation problem has direct applicability in strategies of immunization in disease spreading processes, and influence maximization for certain classes of opinion dynamical models. In this paper, we consider the problem of optimal percolation on multiplex networks. The multiplex scenario serves to realistically model various technological, biological, and social networks. We find that the multilayer nature of these systems, and more precisely multiplex characteristics such as edge overlap and interlayer degree-degree correlation, profoundly changes the properties of the set of nodes identified as the solution of the optimal percolation problem.

physics.soc-ph↗

arXiv2017

Redundant interdependencies boost the robustness of multilayer networks

In the analysis of the robustness of multiplex networks, it is commonly assumed that a node is functioning only if its interdependent nodes are simultaneously functioning. According to this model, a multiplex network becomes more and more fragile as the number of layers increases. In this respect, the addition of a new layer of interdependent nodes to a preexisting multiplex network will never improve its robustness. Whereas such a model seems appropriate to understand the effect of interdependencies in the simplest scenario of a network composed of only two layers, it may seem not suitable to characterize the robustness of real systems formed by multiple network layers. It seems in fact unrealistic that a real system, evolved, through the development of multiple layers of interactions, towards a fragile structure. In this paper, we introduce a model of percolation where the condition that makes a node functional is that the node is functioning in at least two of the layers of the network. The model reduces to the commonly adopted percolation model for multiplex networks when the number of layers equals two. For larger number of layers, however, the model describes a scenario where the addition of new layers boosts the robustness of the system by creating redundant interdependencies among layers. We prove this fact thanks to the development of a message-passing theory able to characterize the model in both synthetic and real-world multiplex graphs.

physics.soc-ph↗

arXiv2017

Fundamental difference between superblockers and superspreaders in networks

Two very important problems regarding spreading phenomena in complex topologies are the optimal selection of node sets either to minimize or maximize the extent of outbreaks. Both problems are nontrivial when a small fraction of the nodes in the network can be used to achieve the desired goal. The minimization problem is equivalent to a structural optimization. The "superblockers", i.e., the nodes that should be removed from the network to minimize the size of outbreaks, are those nodes that make connected components as small as possible. "Superspreaders" are instead the nodes such that, if chosen as initiators, they maximize the average size of outbreaks. The identity of superspreaders is expected to depend not just on the topology, but also on the specific dynamics considered. Recently, it has been conjectured that the two optimization problems might be equivalent, in the sense that superblockers act also as superspreaders. In spite of its potential groundbreaking importance, no empirical study has been performed to validate this conjecture. In this paper, we perform an extensive analysis over a large set of real-world networks to test the similarity between sets of superblockers and of superspreaders. We show that the two optimization problems are not equivalent: superblockers do not act as optimal spreaders.

physics.soc-ph↗

arXiv2017

Observability transition in multiplex networks

We extend the observability model to multiplex networks. We present mathematical frameworks, valid under the treelike ansatz, able to describe the emergence of the macroscopic cluster of mutually observable nodes in both synthetic and real-world multiplex networks. We show that the observability transition in synthetic multiplex networks is discontinuous. In real-world multiplex networks instead, edge overlap among layers is responsible for the disappearance of any sign of abruptness in the emergence of the the macroscopic cluster of mutually observable nodes.

physics.soc-ph↗

arXiv2016

Citation success index - An intuitive pair-wise journal comparison metric

In this paper we present "citation success index", a metric for comparing the citation capacity of pairs of journals. Citation success index is the probability that a random paper in one journal has more citations than a random paper in another journal (50% means the two journals do equally well). Unlike the journal impact factor (IF), the citation success index depends on the broadness and the shape of citation distributions. Also, it is insensitive to sporadic highly-cited papers that skew the IF. Nevertheless, we show, based on 16,000 journals containing ~2.4 million articles, that the citation success index is a relatively tight function of the ratio of IFs of journals being compared, due to the fact that journals with same IF have quite similar citation distributions. The citation success index grows slowly as a function of IF ratio. It is substantial (>90%) only when the ratio of IFs exceeds ~6, whereas a factor of two difference in IF values translates into a modest advantage for the journal with higher IF (index of ~70%). We facilitate the wider adoption of this metric by providing an online calculator that takes as input parameters only the IFs of the pair of journals.

cs.DL↗

arXiv2016

Quantifying perceived impact of scientific publications

Citations are commonly held to represent scientific impact. To date, however, there is no empirical evidence in support of this postulate that is central to research assessment exercises and Science of Science studies. Here, we report on the first empirical verification of the degree to which citation numbers represent scientific impact as it is actually perceived by experts in their respective field. We run a large-scale survey of about 2000 corresponding authors who performed a pairwise impact assessment task across more than 20000 scientific articles. Results of the survey show that citation data and perceived impact do not align well, unless one properly accounts for strong psychological biases that affect the opinions of experts with respect to their own papers vs. those of others. First, researchers tend to largely prefer their own publications to the most cited papers in their field of research. Second, there is only a mild positive correlation between the number of citations of top-cited papers in given research areas and expert preference in pairwise comparisons. This also applies to pairs of papers with several orders of magnitude differences in their total number of accumulated citations. However, when researchers were asked to choose among pairs of their own papers, thus eliminating the bias favouring one's own papers over those of others, they did systematically prefer the most cited article. We conclude that, when scientists have full information and are making unbiased choices, expert opinion on impact is congruent with citation numbers.

cs.DL↗

arXiv2016

Percolation in real multiplex networks

We present an exact mathematical framework able to describe site-percolation transitions in real multiplex networks. Specifically, we consider the average percolation diagram valid over an infinite number of random configurations where nodes are present in the system with given probability. The approach relies on the locally treelike ansatz, so that it is expected to accurately reproduce the true percolation diagram of sparse multiplex networks with negligible number of short loops. The performance of our theory is tested in social, biological, and transportation multiplex graphs. When compared against previously introduced methods, we observe improvements in the prediction of the percolation diagrams in all networks analyzed. Results from our method confirm previous claims about the robustness of real multiplex networks, in the sense that the average connectedness of the system does not exhibit any significant abrupt change as its individual components are randomly destroyed.

physics.soc-ph↗

arXiv2016

Observability transition in real networks

We consider the observability model in networks with arbitrary topologies. We introduce a system of coupled nonlinear equations, valid under the locally tree-like ansatz, to describe the size of the largest observable cluster as a function of the fraction of directly observable nodes present in the network. We perform a systematic analysis on 95 real-world graphs and compare our theoretical predictions with numerical simulations of the observability model. Our method provides almost perfect predictions in the majority of the cases, even for networks with very large values of the clustering coefficient. Potential applications of our theory include the development of efficient and scalable algorithms for real-time surveillance of social networks, and monitoring of technological networks.

physics.soc-ph↗

arXiv2016

Breaking of the site-bond percolation universality in networks

The stochastic addition of either vertices or connections in a network leads to the observation of the percolation transition, a structural change with the appearance of a connected component encompassing a finite fraction of the system. Percolation has always been regarded as a substrate-dependent but model-independent process, in the sense that the critical exponents of the transition are determined by the geometry of the system, but they are identical for the bond and site percolation models. Here, we report a violation of such assumption. We provide analytical and numerical evidence of a difference in the values of the critical exponents between the bond and site percolation models in networks with null percolation thresholds, such as scale-free graphs with diverging second moment of the degree distribution. We discuss possible implications of our results in real networks, and provide additional insights on the anomalous nature of the percolation transition with null threshold.

physics.soc-ph↗

arXiv2016

Leveraging percolation theory to single out influential spreaders in networks

Among the consequences of the disordered interaction topology underlying many social, techno- logical and biological systems, a particularly important one is that some nodes, just because of their position in the network, may have a disproportionate effect on dynamical processes mediated by the complex interaction pattern. For example, the early adoption by an opinion leader in a social network may change the fate of a commercial product, or just a few super-spreaders may determine the virality of a meme in social media. Despite many recent efforts, the formulation of an accurate method to optimally identify influential nodes in complex network topologies remains an unsolved challenge. Here, we present the exact solution of the problem for the specific, but highly relevant, case of the Susceptible-Infected-Removed (SIR) model for epidemic spreading at criticality. By exploiting the mapping between bond percolation and the static properties of SIR, we prove that the recently introduced Non-Backtracking centrality is the optimal criterion for the identification of influential spreaders in locally tree-like networks at criticality. By means of simulations on synthetic networks and on a very extensive set of real-world networks, we show that the Non-Backtracking centrality is a highly reliable metric to identify top influential spreaders also in generic graphs not embedded in space, and for noncritical spreading.

physics.soc-ph↗

arXiv2016

Beyond the locally tree-like approximation for percolation on real networks

Theoretical attempts proposed so far to describe ordinary percolation processes on real-world networks rely on the locally tree-like ansatz. Such an approximation, however, holds only to a limited extent, as real graphs are often characterized by high frequencies of short loops. We present here a theoretical framework able to overcome such a limitation for the case of site percolation. Our method is based on a message passing algorithm that discounts redundant paths along triangles in the graph. We systematically test the approach on 98 real-world graphs and on synthetic networks. We find excellent accuracy in the prediction of the whole percolation diagram, with significant improvement with respect to the prediction obtained under the locally tree-like approximation. Residual discrepancies between theory and simulations do not depend on clustering and can be attributed to the presence of loops longer than three edges. We present also a method to account for clustering in bond percolation, but the improvement with respect to the method based on the tree-like approximation is much less apparent.

physics.soc-ph↗

arXiv2015

Defining and identifying Sleeping Beauties in science

A Sleeping Beauty (SB) in science refers to a paper whose importance is not recognized for several years after publication. Its citation history exhibits a long hibernation period followed by a sudden spike of popularity. Previous studies suggest a relative scarcity of SBs. The reliability of this conclusion is, however, heavily dependent on identification methods based on arbitrary threshold parameters for sleeping time and number of citations, applied to small or monodisciplinary bibliographic datasets. Here we present a systematic, large-scale, and multidisciplinary analysis of the SB phenomenon in science. We introduce a parameter-free measure that quantifies the extent to which a specific paper can be considered an SB. We apply our method to 22 million scientific papers published in all disciplines of natural and social sciences over a time span longer than a century. Our results reveal that the SB phenomenon is not exceptional. There is a continuous spectrum of delayed recognition where both the hibernation period and the awakening intensity are taken into account. Although many cases of SBs can be identified by looking at monodisciplinary bibliographic data, the SB phenomenon becomes much more apparent with the analysis of multidisciplinary datasets, where we can observe many examples of papers achieving delayed yet exceptional importance in disciplines different from those where they were originally published. Our analysis emphasizes a complex feature of citation dynamics that so far has received little attention, and also provides empirical evidence against the use of short-term citation metrics in the quantification of scientific impact.

physics.soc-ph↗

arXiv2015

Percolation in real interdependent networks

The function of a real network depends not only on the reliability of its own components, but is affected also by the simultaneous operation of other real networks coupled with it. Robustness of systems composed of interdependent network layers has been extensively studied in recent years. However, the theoretical frameworks developed so far apply only to special models in the limit of infinite sizes. These methods are therefore of little help in practical contexts, given that real interconnected networks have finite size and their structures are generally not compatible with those of graph toy models. Here, we introduce a theoretical method that takes as inputs the adjacency matrices of the layers to draw the entire phase diagram for the interconnected network, without the need of actually simulating any percolation process. We demonstrate that percolation transitions in arbitrary interdependent networks can be understood by decomposing these system into uncoupled graphs: the intersection among the layers, and the remainders of the layers. When the intersection dominates the remainders, an interconnected network undergoes a continuous percolation transition. Conversely, if the intersection is dominated by the contribution of the remainders, the transition becomes abrupt even in systems of finite size. We provide examples of real systems that have developed interdependent networks sharing a core of "high quality" edges to prevent catastrophic failures.

physics.soc-ph↗

arXiv2015

Predicting percolation thresholds in networks

We consider different methods, that do not rely on numerical simulations of the percolation process, to approximate percolation thresholds in networks. We perform a systematic analysis on synthetic graphs and a collection of 109 real networks to quantify their effectiveness and reliability as prediction tools. Our study reveals that the inverse of the largest eigenvalue of the non-backtracking matrix of the graph often provides a tight lower bound for true percolation threshold. However, in more than 40% of the cases, this indicator is less predictive than the naive expectation value based solely on the moments of the degree distribution. We find that the performance of all indicators becomes worse as the value of the true percolation threshold grows. Thus, none of them represents a good proxy for robustness of extremely fragile networks.

physics.soc-ph↗

arXiv2014

Quality versus quantity in scientific impact

Citation metrics are becoming pervasive in the quantitative evaluation of scholars, journals and institutions. More then ever before, hiring, promotion, and funding decisions rely on a variety of impact metrics that cannot disentangle quality from quantity of scientific output, and are biased by factors such as discipline and academic age. Biases affecting the evaluation of single papers are compounded when one aggregates citation-based metrics across an entire publication record. It is not trivial to compare the quality of two scholars that during their careers have published at different rates in different disciplines in different periods of time. We propose a novel solution based on the generation of a statistical baseline specifically tailored on the academic profile of each researcher. Our method can decouple the roles of quantity and quality of publications to explain how a certain level of impact is achieved. The method is flexible enough to allow for the evaluation of, and fair comparison among, arbitrary collections of papers --- scholar publication records, journals, and entire institutions; and can be extended to simultaneously suppresses any source of bias. We show that our method can capture the quality of the work of Nobel laureates irrespective of number of publications, academic age, and discipline, even when traditional metrics indicate low impact in absolute terms. We further apply our methodology to almost a million scholars and over six thousand journals to measure the impact that cannot be explained by the volume of publications alone.

cs.DL↗

arXiv2014

Underestimating extreme events in power-law behavior due to machine-dependent cutoffs

Power-law distributions are typical macroscopic features occurring in almost all complex systems observable in nature. As a result, researchers in quantitative analyses must often generate random synthetic variates obeying power-law distributions. The task is usually performed through standard methods that map uniform random variates into the desired probability space. Whereas all these algorithms are theoretically solid, in this paper we show that they are subject to severe machine-dependent limitations. As a result, two dramatic consequences arise: (i) the sampling in the tail of the distribution is not random but deterministic; (ii) the moments of the sample distribution, which are theoretically expected to diverge as functions of the sample sizes, converge instead to finite values. We provide quantitative indications for the range of distribution parameters that can be safely handled by standard libraries used in computational analyses. Whereas our findings indicate possible reinterpretations of numerical results obtained through flawed sampling methodologies, they also pave the way for the search for a concrete solution to this central issue shared by all quantitative sciences dealing with complexity.

physics.soc-ph↗

arXiv2014

A paradox in community detection

Recent research has shown that virtually all algorithms aimed at the identification of communities in networks are affected by the same main limitation: the impossibility to detect communities, even when these are well-defined, if the average value of the difference between internal and external node degrees does not exceed a strictly positive value, in literature known as detectability threshold. Here, we counterintuitively show that the value of this threshold is inversely proportional to the intrinsic quality of communities: the detection of well-defined modules is thus more difficult than the identification of ill-defined communities.

physics.soc-ph↗