Cores in temporal networks A longitudinal approach

20
Cores in temporal networks A longitudinal approach Vladimir Batagelj Institute of Mathematics, Physics and Mechanics, Jadranska 19, 1000 Ljubljana, Slovenia, IAM, University of Primorska, 6000 Koper, Slovenia, ANR Lab High School of Economics, Moscow, Russia Monika Cerinšek Abelium d.o.o. R&D, Kajuhova 90, 1000 Ljubljana, Slovenia November 7, 2021 Abstract Among many formalizations of the notion of dense subnetworks of a given network (clique, s-plexes, LS sets, lambda sets, cores, etc.) only for cores exists an efficient algorithm that can compute the result in an acceptable time also for large networks. In a graph G =(V,E) a subgraph H =(C, E(C)) induced by the subset of nodes C is a k-core or a core of order k iff for each node v in C its internal degree deg H (v) is greater or equal to k, and H is a maximal subgraph with this property. Various generalizations of ordinary cores according to types of net- works and measures of node importance were proposed. In our contribu- tion we extend the notion of cores to temporal networks and present the corresponding algorithms. For illustration we present results of determin- ing cores in selected artificial and real-life networks. Keywords: large network, temporal network, core, efficient algorithm, temporal quantity Mathematics Subject Classification: 91D30; 05C22; 05C82; 05C85; 62R07; 1 Introduction There are many formalizations of the notion of dense subnetwork(s) of a given network (cliques, s-plexes, LS sets, lambda sets, cores, etc. [22]). In most cases the corresponding algorithms are too expensive (require too much time) to be applied to large networks. A very efficient algorithms exists for determining cores. Besides this, some other interesting types of subnetworks are contained 1

Transcript of Cores in temporal networks A longitudinal approach

Page 1: Cores in temporal networks A longitudinal approach

Cores in temporal networksA longitudinal approach

Vladimir Batagelj

Institute of Mathematics, Physics and Mechanics, Jadranska 19,1000 Ljubljana, Slovenia,

IAM, University of Primorska, 6000 Koper, Slovenia,ANR Lab High School of Economics, Moscow, Russia

Monika Cerinšek

Abelium d.o.o. R&D, Kajuhova 90, 1000 Ljubljana, Slovenia

November 7, 2021

AbstractAmong many formalizations of the notion of dense subnetworks of a

given network (clique, s-plexes, LS sets, lambda sets, cores, etc.) onlyfor cores exists an efficient algorithm that can compute the result in anacceptable time also for large networks.

In a graph G = (V, E) a subgraph H = (C, E(C)) induced by thesubset of nodes C is a k-core or a core of order k iff for each node v in Cits internal degree degH(v) is greater or equal to k, and H is a maximalsubgraph with this property.

Various generalizations of ordinary cores according to types of net-works and measures of node importance were proposed. In our contribu-tion we extend the notion of cores to temporal networks and present thecorresponding algorithms. For illustration we present results of determin-ing cores in selected artificial and real-life networks.Keywords: large network, temporal network, core, efficient algorithm,temporal quantityMathematics Subject Classification: 91D30; 05C22; 05C82; 05C85;62R07;

1 IntroductionThere are many formalizations of the notion of dense subnetwork(s) of a givennetwork (cliques, s-plexes, LS sets, lambda sets, cores, etc. [22]). In most casesthe corresponding algorithms are too expensive (require too much time) to beapplied to large networks. A very efficient algorithms exists for determiningcores. Besides this, some other interesting types of subnetworks are contained

1

Page 2: Cores in temporal networks A longitudinal approach

in cores – cores can be used for efficient localization of such subnetworks [6, ].This makes cores very attractive and useful.

The notion of k-core was introduced in 1983 by Seidman[20]. A linear innumber of links algorithm was presented by Batagelj and Zaveršnik in 2003[9, 11, 6] and extended to other node property functions beside degrees [10, 6].In [1, 13] the (p, q)-cores were proposed as an extension of the idea to two-modenetworks. In this paper we present a further extension to temporal networksbased on the notion of temporal quantities [5].

Different ideas and applications of cores were exaustively reviewed in a recentpaper [18]. There are some recent works on temporal cores that can be classifiedin two groups: (1) core maintenance in networks changing by a stream of events[17, 23, 24, 4, 3, 21, 24, 2]; (2) the notion of a span-core [16].

Our approach differs because we use for each node a temporal quantitydescribing changes of its core number through time. In the paper we extend thenotion of a p-core to temporal networks and present the algorithms to determinethe temporal cores based on degrees and weighted degrees.

In Section 2 we present definitions of basic notions used in our method(temporal network, temporal core). In Section 3 we present the algorithm forour method. In Section 4 some results obtained in usage of our method ofartificial and real-life data are presented.

2 Definitions2.1 CoresA network N = (V,L,P,W); n = |V|, m = |L| is based on four sets: the set ofnodes V, the set of links L, the set of node properties P, and the set of weightsW. Each link is linking two nodes – its end-nodes. A link is either directed –an arc, or undirected – an edge. A pair of sets (V.L) forms a graph.

The notion of a k-core was introduced by Seidman in 1983 [20]: Let G =(V,L) be a graph with set of nodes V and set of links L. The degree of nodev ∈ V is denoted by deg(v). For a given integer k, a subgraph Hk = (Ck,L|Ck)induced by the subset of nodes Ck ⊆ V is called a k-core or a core of order k iffdegHk

(v) ≥ k, for all v ∈ Ck, and Hk is the maximum such subgraph.The core of maximum order is called the main core. The core number ,

core(v), of node v is the highest order of a core that contains this node.Let N(v) denote the set of neighbors of a node v in a network N , and

N(v, C) = N(v) ∩ C is the set of neighbors of a node v within the subset ofnodes C ⊆ V.

A very efficient algorithm exists for cores. All nodes have degree at least0. They belong to the core H0 of order 0. Removing all nodes of degree 0we obtain the set C1 in which each node has degree at least 1 – the core H1of order 1. Removing all nodes of (current) degree 1 we obtain the set C2 inwhich each node has degree at least 2 – the core H2 of order 2. . . . This leadsto an algorithm for determining each node’s core number core[v] presented inAlgorithm 1. An efficient, O(|L|), implementation of this algorithm is given in[6].

2

Page 3: Cores in temporal networks A longitudinal approach

Algorithm 1: Core decomposition algorithm.

1 CoreDecomposition (N ) :2 deg = {u : deg(u) for u ∈ V}3 C = V ; k = 1 ; core = { }4 while ¬empty(C) :5 while ∃u ∈ C : deg[u] < k :6 for v ∈ N(u, C) :7 deg[v] = deg[v]− 18 C = C \ {u}9 core[u] = k − 1

10 k = k + 111 return core

2.2 Generalized coresAssume that in a network N = (V,L,P,W) a node property function p(v, C) isdefined, p : V ×2V → R, where C ⊆ V is a cluster – a subset of nodes, and v ∈ Cis a node.

Some examples of such node properties were proposed in [6]:

1. p1(v, C) = degC(v): node degree within the induced subgraph N (v, C) =(C,L(C));

2. p2(v, C) = indegC(v)+outdegC(v): if all links are directed it holds p2 = p1;

3. p3(v, C) = wdeg(v, C) =∑

u∈N(v,C) w(v, u) for w : L → R+0 : weighted

degree, the sum of weights of incident links within N (v, C);

4. p4(v, C) = maxu∈N(v,C) w(v, u) for w : L → R: the maximum weight ofincident links within N (v, C);

5. p5(v, C) = degC(v)deg(v) if deg(v) > 0 else f5(v, C) = 0: the fraction of neighbors

within N (v, C);

6. p6(v, C) = wdeg(v,C)wdeg(v,V) for w : L → R+

0 : the fraction of the weighted degreeof incident links within N (v, C).

The node property functions are used in the definition of generalized cores[6].

The subgraph H = (C,L(C)) induced by the set C ⊆ V is a p-core at levelt ∈ R iff ∀v ∈ C : t ≤ p(v, C) and C is the maximum such set.

We say that the node property function p(v, C):

• is local iff: p(v, C) = p(v, N(v, C)) ∀v ∈ V.

• is monotonic iff: C1 ⊂ C2 ⇒ ∀v ∈ V : p(v, C1) ≤ p(v, C2)..

For a local and monotone property function p the corresponding p-core numberscan be efficiently determined using a generalized version of Algorithm 1 [6].

3

Page 4: Cores in temporal networks A longitudinal approach

2.3 Temporal network and temporal coreWe get a temporal network NT = (V,L, T ,P,W) by attaching the time T toan ordinary network, where T = [Tmin, Tmax) is a linearly ordered set of timepoints t ∈ T which are usually integers or reals. In a temporal network activity(presence / absence) of nodes and links may change through time. Let TV (v),TV ∈ P, be the activity set of time points for node v ∈ V and TL(`), TL ∈ W,the activity set of time points for link ` ∈ L. Consistency condition must befulfilled for activity sets: If a link `(u, v) is active at the time point t then itsend-nodes u and v should be active at the time point t :

TL(`(u, v)) ⊆ TV (u) ∩ TV (v).

Beside this also the values of node properties and link weights may change.These changes can be described in different ways. In this paper we will use theapproach based on temporal quantities (TQs) [5, 8].

A temporal quantity a with the activity set Ta ⊆ T describes the changes ofits value through time:

a ={

a(t) t ∈ Ta;t ∈ T \Ta.

We assume that the values of temporal quantities belong to a set A which is asemiring (A, +, ·, 0, 1) for binary operations + : A × A → A and · : A × A →A. In this paper we will limit our attention to the combinatorial semiring(R+

0 , +, ·, 0, 1) where + is the addition and · is the multiplication of numbers.We can extend both operations to the set A = A ∪ { } by requiring that forall a ∈ A it holds

a + = + a = a and a · = · a =

The structure (A , +.·, , 1) is also a semiring. Let us denote A (T ) the set ofall temporal quantities over A in time T . To extend the operations to networksand their matrices we first define the sum (parallel links) a + b as

(a + b)(t) = a(t) + b(t) and Ta+b = Ta ∪ Tb.

and the product (sequential links) a · b as

(a · b)(t) = a(t) · b(t) and Ta·b = Ta ∩ Tb.

Nodes and links in temporal networks have temporal quantities assignedas values of some properties/weights. For cumputational reasons we limit ourfurther discussion to temporal quantities or TQs that are represented with alist of triples (s, f, v) and each triple determine the time interval [s, f) in whichnode or link is active or present, and its value v is constant on this time interval.

An example of temporal network is shown in Figure 1. It is defined on adiscrete time set T = [1, 9) = {1, 2, 3, 4, 5, 6, 7, 8}. All nodes in this networkare present through whole time interval. Some links are directed – direction isindicated with an arrow. All links have value 1 when they are present. Somelinks are not present through all the time. For example, link from node 2 tonode 4 is present in time points [3, 9) = {3, 4, 5, 6, 7, 8}.

4

Page 5: Cores in temporal networks A longitudinal approach

A time slice N (t) = (V(t),L(t),P(t),W(t)) of the temporal network NT

in a time point t consists of nodes and links active in a time point t and thecorresponding values of node properties and link weights in a time point t. Thenotion of a time slice can be extended to a time interval I ⊂ T by

N (I) =⋃t∈IN (t)

The p-core number p-core(v, t) of node v ∈ V in time point t ∈ TV (v) isequal to the p-core number of node v in the time slice N (t); and is equal tootherwise. It determines a TQ p-core(v).

3 Algorithms3.1 Temporal degree coresThe algorithm for temporal core decomposition is based on algorithm for k-coredecomposition that is presented in Alg. 1.

In the description of the algorithm we will use Python-like comprehensionexpressions for lists

L = [E(s) for s ∈ S if P (s)]

and for dictionaries

D = {s : E(s) for s ∈ S if P (s)}

S is a sequence (list, set, dictionary), s is an item from the sequence; E(s) is anexpression transforming the item s into a value.

• TQsum(a, b) – returns the sum a + b of TQs a and b;

• TQminus(a) – returns the TQ −a;

• TQsetConst(a, k) – sets the value component in each triple of the TQ ato k;

• TQmin(a) – returns minimum of values in triples of the TQ a; returns ∞

• TQcutGE(a, k) – removes from the TQ a all triples tq with tq.v < k;

• TQcutGT (a, k) – removes from the TQ a all triples tq with tq.v ≤ k;

• TQcutLE(a, k) – removes from the TQ a all triples tq with tq.v > k;

• TQcutEQ(a, k) – removes from the TQ a all triples tq with tq.v 6= k;

• TQlower(a, k) – returns the TQ a in which values smaller than k are setto k;

• TQextract(a, b) – extracts a from b – restricts the TQ b to the activityset Ta of the TQ a;

• TQcomplement(a, tmin, tmax) – returns the TQ with triples determinedby the activity set [tmin, tmax)\Ta with the value component equal to 1;

5

Page 6: Cores in temporal networks A longitudinal approach

• TQdeg(v) – returns the temporal degree of the node v ∈ V;

• TQwdeg(v) – returns the temporal weighted degree of the node v ∈ V

wdeg(v) =∑

`∈star(v)

w(`)

D is a dictionary of (still active) node temporal degrees. D[u] is a temporaldegree of the node u ∈ V. Core is a dictionary of currently known temporalcore numbers. Core[u] is a TQ with temporal core numbers of the node u ∈ V.

tq is a triple from a TQ. It has components: tq.s – start, tq.f – finish, andtq.v – value.

Algorithm 2: Degree cores in temporal network.

1 TemporalDegreeCores (N ) :2 D = {u : TQdeg(u) for u ∈ V}3 Core = {u : TQcutLE(D[u], 0) for u ∈ V}4 D = {u : TQcutGT (D[u], 0) for u ∈ V}5 D = {u : D[u] for u ∈ V i f ¬empty(D[u])}6 Dmin = {u : TQmin(D[u]) for u ∈ D}7 while ¬empty(D) :8 u = argmin(Dmin) ; dmin = Dmin[u]9 core = [tq for tq ∈ D[u] i f tq.v = dmin]

10 Core[u] = TQsum(Core[u], core)11 change = TQsetConst(core,−1)12 D[u] = TQcutGE(TQsum(D[u], change), dmin)13 i f empty(D[u]) : delete(D, u) ; delete(Dmin, u)14 else : Dmin[u] = TQmin(D[u])15 for ` in star(u) :16 v = twin(u, `)17 i f v /∈ D : continue18 chLink = TQextract(w(`), change)19 i f empty(chLink) : continue20 diff = TQcutGE(TQsum(D[v], chLink), 0)21 i f empty(diff) : delete(D, v) ; delete(Dmin, v)22 else :23 D[v] = TQlower(diff, dmin)24 Dmin[v] = TQmin(D[v])25 return Core

DELETE ???We expanded Alg. 1 onto temporal quantities and the new algorithm is shown

in Alg. 3. Input for this algorithm is temporal network. Before main loop wecopy node temporal quantities into new set D, define set of core numbers fornode in shape of temporal quantities, removed temporal quantities from D thathave a degree value equal to 0, and defined the dictionary Dmin which consistsof keys equal to nodes and values equal to minimum degree of a node. This isshown in lines 2-5 in Alg. 3.

6

Page 7: Cores in temporal networks A longitudinal approach

The main loop (lines 6-29 in Alg. 3) is repeated until the set D is not empty.At first we find a pair (dmin, u) from Dmin where dmin is minimum of all valuesin Dmin. As core we define set of temporal quantities from D[u] with degreeequal to dmin. We add these temporal quantities to the set of core numbers(line 11 in Alg. 3) as their degree value represent core numbers for node u ingiven time intervals. We rewrite core into change with degree values equal to−1. This we need to save only those temporal quantities in D[u] that have adegree larger than dmin where we reduce their degrees for 1 in those that arealso in change.

Now we can check all neighbours of node u – we loop through all links ofu – lines 14-25 in Alg. 3. If the other end-node of a link is not in D, we skipit and check another link. Otherwise we define changeLink as intersection oftemporal quantities of a link and change : time intervals are intersections of timeintervals, value is equal to −1. Link changeLink is actually active the same timeas u with degree value equal to dmin. If this link is never active, we skip it andcheck another link neighbouring u. Now we define diff as temporal quantitiesfrom D[v] with degree value more than 0, where we reduce their degrees for 1in those that are also in changeLink. Now we rewrite D[v] as diff with degreevalues equal to max(deg, dmin). If D[v] is empty, we delete node v from D andDmin. Otherwise we set a degree value for Dmin[v] to be equal to minimumdegree value in D[v].

After checking all neighbouring links of node u we check if there are anynon-empty temporal quantities in D[u] – lines 26-29 in Alg. 3. If there are none,we delete node u from D and Dmin. Otherwise we set a degree value for u inDmin to be minimum degree value in temporal quantities of D[u].

*** Correctness check an instance t

3.1.1 Example

The network is shown in Fig. ?? Figure 5 from [5]. It consists of 15 nodes thatare always present. Most of the links are directed, some are undirected. Mostof the links are present all the time T = [1, 9) = {1, 2, 3, 4, 5, 6, 7, 8}, Tmin = 1and Tmax = 9 – for them there are no temporal quantities written in figure.Links (2, 4), (5, 7), (11, 7), (13, 14), (13, 15), (14, 15) are present in defines timeintervals. All links have value equal to 1 when they are active. The degrees ofnodes are listed in the left side of Tab. 1. For example, node 13 has a degreeequal to 0 in the time interval [1, 2) (time point 1) because both neighbouringlinks are not present in this time interval. In time interval [2, 8) are both linkspresent, so the node has a degree equal to 2. After that its degree is again equalto 0, because no incident link is active.

In the algorithm for temporal core decomposition we need to define sets D,CoreHierarchy, and Dmin. For this examples these are:

• D = {1 : [(1, 9, 1)], 2 : [(1, 3, 2), (3, 9, 3)], ..., 15 : [(1, 2, 0), (2, 8, 2), (8, 9, 0)]}

• CoreHierarchy = {12 : [(1, 9, 0)], 13 : [(1, 2, 0), (8, 9, 0)], 14 : [(1, 2, 0),(8, 9, 0)], 15 : [(1, 2, 0), (8, 9, 0)]}

• Remove temporal quantities with value 0 from D: D = {1 : [(1, 9, 1)], 2 :[(1, 3, 2), (3, 9, 3)], ..., 15 : [(2, 8, 2)]}

7

Page 8: Cores in temporal networks A longitudinal approach

Figure 1: First example network. All unlabeled links have a value of [(1, 9, 1)].

• Dmin = {1 : 1, 2 : 2, 3 : 1, 4 : 2, 5 : 2, 6 : 2, 7 : 3, 8 : 4, 9 : 4, 10 : 4, 11 :3, 12 : 0, 13 : 0, 14 : 0, 15 : 0}

We determine dmin from Dmin to be 0, which is first core number wecalculate cores with. The result of the algorithm is shown on the right in Tab. 1.Core number for node 1 is equal 1 all the time. Core number for node 15 isequal to 0 in time intervals [1, 2) and [8, 9), and is equal to 2 in time interval[2, 8).

3.2 Temporal weighted degree corespS-cores are very similar to ordinary cores. For ordinary cores the node propertyfunction is p1 = deg, and for pS-cores the node property function is p3 = wdeg.Note that in the case w(`) = 1 for all ` ∈ L p3 = wdeg = deg = p1. Thereforethe pS-cores algorithm can be used also for computing ordinary cores.

We can base our algorithm for pS-cores on Alg.2 with some adaptations forchanges of the values of property function.

In application of Algorithm 3 to some bibliographic networks based on frac-tional approach [] it turned out that it can have problems with rounding errors(*** example). To account for them we introduced the precision parameter eps(for example, eps = 10−5) and replace lines 10 and 19 with

10 core = TQextract(cCore, [tq for tq ∈ D[u] i f tq.v ≤ dmin + eps])

and

19 diff = TQcutGE(TQsum(D[v], chLink),−eps)

8

Page 9: Cores in temporal networks A longitudinal approach

Table 1: Degrees and degree core numbers for first artificial network.

Node Degree Degree core number1 (1, 9, 1) (1, 9, 1)2 (1, 3, 2), (3, 9, 3) (1, 9, 1)3 (1, 9, 1) (1, 9, 1)4 (1, 3, 2), (3, 9, 3) (1, 9, 2)5 (1, 5, 3), (5, 9, 2) (1, 9, 2)6 (1, 9, 2) (1, 9, 2)7 (1, 5, 4), (5, 7, 3), (7, 9, 4) (1, 7, 3), (7, 9, 4)8 (1, 9, 4) (1, 7, 3), (7, 9, 4)9 (1, 9, 4) (1, 7, 3), (7, 9, 4)

10 (1, 9, 4) (1, 7, 3), (7, 9, 4)11 (1, 7, 3), (7, 9, 4) (1, 7, 3), (7, 9, 4)12 (1, 9, 0) (1, 9, 0)13 (1, 2, 0), (2, 8, 2), (8, 9, 0) (1, 2, 0), (2, 8, 2), (8, 9, 0)14 (1, 2, 0), (2, 8, 2), (8, 9, 0) (1, 2, 0), (2, 8, 2), (8, 9, 0)15 (1, 2, 0), (2, 8, 2), (8, 9, 0) (1, 2, 0), (2, 8, 2), (8, 9, 0)

Algorithm 3: Weighted degree cores in temporal network.

1 TemporalWeightedDegreeCores (N ) :2 D = {u : TQwdeg(u) for u ∈ V}3 Core = {u : TQcutLE(D[u], 0) for u ∈ V}4 D = {u : TQcutGT (D[u], 0) for u ∈ V}5 D = {u : D[u] for u ∈ V i f ¬empty(D[u])}6 Dmin = {u : TQmin(D[u]) for u ∈ D}7 while ¬empty(D) :8 u = argmin(Dmin) ; dmin = Dmin[u]9 cCore = TQcomplement(Core[u], Tmin, Tmax)

10 core = TQextract(cCore, TQcutLE(D[u], dmin))11 i f ¬empty(core) :12 Core[u] = TQsum(Core[u], core)13 D[u] = TQcutGE(TQsum(D[u], TQminus(core)), dmin)14 for ` in star(u) :15 v = twin(u, `)16 i f v /∈ D : continue17 chLink = TQminus(TQextract(core, w(`)))18 i f empty(chLink) : continue19 diff = TQcutGE(TQsum(D[v], chLink), 0)20 D[v] = TQlower(diff, dmin)21 i f empty(D[v]) : delete(D, v) ; delete(Dmin, v)22 else : Dmin[v] = TQmin(D[v])23 i f empty(D[u]) : delete(D, u) ; delete(Dmin, u)24 else : Dmin[u] = TQmin(D[u])25 return Core

9

Page 10: Cores in temporal networks A longitudinal approach

3.2.1 Example

We defined another artificial temporal network (Fig. 2) with the same underlyinggraph structure as in previous network (Fig. ??). The presence and values oflinks are different which is shown in Fig. 2 and in values of nodes in Tab. 2. Weselected sum of neighbouring link values as node values for this example so wecan calculate pS-core numbers for nodes. The result of the algorithm is shownon the right side of Tab. 2.

Figure 2: Visualization of second artificial example of a temporal network. Asingle value v on an arc is an abreviation for [(1, 9, v)]

4 ResultsWe tested our method on artificial examples and examples from real-life withdifferent context. Firstly we will look at artificial example to better understandwhat temporal network is and how do we determine temporal cores.

4.1 Reuters terror news networkThe Reuters terror news network was obtained using the CRA (Centering Reso-nance Analysis) by Steve Corman and Kevin Dooley at Arizona State University[14]. This temporal network is available at

http://vlado.fmf.uni-lj.si/pub/networks/data/CRA/terror.htm.The source data are all stories released during 66 consecutive days by the

news agency Reuters concerning the September 11 attack on the U.S. This timeinterval starts at 9:00 AM EST 11th of September 2001.

10

Page 11: Cores in temporal networks A longitudinal approach

Table 2: Weighted degrees and weighted degree core numbers for second artificialnetwork.

Node Weighted degree Weighted degree core number1 (1, 5, 3), (5, 9, 5) (1, 5, 3), (5, 9, 5)2 (1, 3, 7), (3, 9, 10) (1, 5, 4), (5, 9, 5)3 (1, 5, 4), (5, 9, 2) (1, 5, 4), (5, 9, 2)4 (1, 3, 4), (3, 9, 7) (1, 5, 4), (5, 9, 5)5 (1, 5, 10), (5, 9, 7) (1, 9, 5)6 (1, 9, 7) (1, 9, 5)7 (1, 5, 13), (5, 7, 10), (7, 9, 14) (1, 9, 10)8 (1, 5, 13), (5, 9, 10) (1, 9, 10)9 (1, 5, 19), (5, 9, 16) (1, 9, 10)

10 (1, 9, 11) (1, 9, 10)11 (1, 7, 11), (7, 9, 15) (1, 9, 10)12 (1, 9, 0) (1, 9, 0)13 (1, 2, 0), (2, 5, 6), (5, 8, 9), (8, 9, 0) (1, 2, 0), (2, 5, 5), (5, 8, 7), (8, 9, 0)14 (1, 2, 0), (2, 8, 7), (8, 9, 0) (1, 2, 0), (2, 5, 5), (5, 8, 7), (8, 9, 0)15 (1, 2, 0), (2, 5, 5), (5, 8, 8), (8, 9, 0) (1, 2, 0), (2, 5, 5), (5, 8, 7), (8, 9, 0)

Nodes of the network are important words / terms. Two terms are linked ifthey co-appear in the same text unit (sentence). The link value is equal to thefrequency of such appearances per day – the network is temporal with a daygranularity. There are n = 13332 nodes and m = 243447 edges in the network.50859 of edges have value larger than 1. There are no loops.

We determined the degree cores in the terror news network. The computa-tion took 1 minute and half on a standard PC. The maximum core number is17. How to present the results? There are 13332 TQs to be inspected.

One approach is to restrict our attention to cores with large core numbers –at least k ∈ N. We produced cuts Ck for k = 6, 10, 12, 15. The sizes (number ofnodes) of the obtained cuts are as follows: |C6| = 2282, |C10| = 494, |C12| = 243,|C15| = 34.

In Figure 3 the TQs from the cut C12 are presented as a heatmap. Theordering of its rows (terms) is determined by a hierarchical clustering, completemethod. For computing the dissimilarity between rows a special measure

d(x, y) = 1|Tx ∪ Ty|

∑t∈Tx∪Ty

|x[t]− y[t]|1 + max(0, x[t], y[t])

was used. In computing d we replace in TQs the undefined value with thevalue −1. Only columns (days) with at least some defined value are presented(Sep 11, Sep 12, Sep 16, Sep 26, Oct 11).

*** comment on core nums 17 and 16The cut C6 has 2282 nodes. Too much to apply the same approach as for

C12. We can still produce a clustering tree (dendrogram, see Figure 4), exportit in PDF, and browse it in a viewer (Marquee Zoom in Acrobat Reader) tovisually detect interesting clusters (subtrees in the dendrogram). We extractthe selected clusters and visualize the corresponding TQs as a heatmap.

*** Comments strongest Sep 26, first six days,

11

Page 12: Cores in temporal networks A longitudinal approach

sep−11

sep−12

sep−16

sep−26

oct−11

terror

attack

new_york

strike

timor

sunday

ship−to−shore

revenge

rehearse

landing

helicopter

east

expectation

casualty

bomb

report

civilian

thursday

stronghold

respond

kandahar

grow

intense

kabul

ultimatum

rule

protester

fresh

embassy

desert

coalition

afghanistan

anti−terror

wednesday

capital

taliban

united_states

pres_bush

fire

set

undergrind

store

stockpile

site

salt

oil

million

mexico

gulf

four

deep

crude

coast

cavern

500

barrel

two

support

spokesman

schedule

saudi

response

reporter

price

possible

place

morgan_stanlly_dw

measure

los_angeles

leader

lawmaker

issue

investigation

interview

include

flag

firefighter

early

district

death

crime

controller

chief

car

assault

ashcroft

a.m.

action

war

victim

unprecedented

tragedy

system

south

service

search

rubble

police

percent

passenger

number

news

national

member

mayor_giuliani

limit

life

information

hijacker

ground

florida

fbi

clear

business

boston

big

bank

authority

act

airline

wreckage

west

trade

top

terrorism

tell

team

target

suspect

survivor

state

source

side

send

section

school

say

return

rescuer

rescue

point

personnel

pennsylvania

operation

officer

night

newspaper

new

minute

mayor

market

man

local

late

investigator

intelligence

inc

group

giant

floor

faa

employee

deadly

damage

control

congress

company

collapse

capitol

brief

bin_laden

airliner

aviation

street

huge

history

explosion

evacuation

cause

devastate

emergency

way

traffic

the_worst

smoke

blood

scene

terrorist

television

military

massive

jet

hospital

home

headquarters

federal

crash

agency

airplane

world_trade_ctr

world

worker

work

white_house

washington

twin

tuesday

tower

time

thousand

security

plane

people

pentagon

part

official

office

nation

morning

manhattan

major

low

large

landmark

hour

hijack

government

flight

financial

day

country

commercial

city

center

call

buildngs

buildng

area

american

america

airport

air

aircraft

12

13

14

15

16

17

sep−11

sep−12

sep−16

sep−26

oct−11

terror

attack

new_york

strike

timor

sunday

ship−to−shore

revenge

rehearse

landing

helicopter

east

expectation

casualty

bomb

report

civilian

thursday

stronghold

respond

kandahar

grow

intense

kabul

ultimatum

rule

protester

fresh

embassy

desert

coalition

afghanistan

anti−terror

wednesday

capital

taliban

united_states

pres_bush

fire

set

undergrind

store

stockpile

site

salt

oil

million

mexico

gulf

four

deep

crude

coast

cavern

500

barrel

two

support

spokesman

schedule

saudi

response

reporter

price

possible

place

morgan_stanlly_dw

measure

los_angeles

leader

lawmaker

issue

investigation

interview

include

flag

firefighter

early

district

death

crime

controller

chief

car

assault

ashcroft

a.m.

action

war

victim

unprecedented

tragedy

system

south

service

search

rubble

police

percent

passenger

number

news

national

member

mayor_giuliani

limit

life

information

hijacker

ground

florida

fbi

clear

business

boston

big

bank

authority

act

airline

wreckage

west

trade

top

terrorism

tell

team

target

suspect

survivor

state

source

side

send

section

school

say

return

rescuer

rescue

point

personnel

pennsylvania

operation

officer

night

newspaper

new

minute

mayor

market

man

local

late

investigator

intelligence

inc

group

giant

floor

faa

employee

deadly

damage

control

congress

company

collapse

capitol

brief

bin_laden

airliner

aviation

street

huge

history

explosion

evacuation

cause

devastate

emergency

way

traffic

the_worst

smoke

blood

scene

terrorist

television

military

massive

jet

hospital

home

headquarters

federal

crash

agency

airplane

world_trade_ctr

world

worker

work

white_house

washington

twin

tuesday

tower

time

thousand

security

plane

people

pentagon

part

official

office

nation

morning

manhattan

major

low

large

landmark

hour

hijack

government

flight

financial

day

country

commercial

city

center

call

buildngs

buildng

area

american

america

airport

air

aircraft

12

13

14

15

16

17

sep−11

sep−12

sep−16

sep−26

oct−11

terror

attack

new_york

strike

timor

sunday

ship−to−shore

revenge

rehearse

landing

helicopter

east

expectation

casualty

bomb

report

civilian

thursday

stronghold

respond

kandahar

grow

intense

kabul

ultimatum

rule

protester

fresh

embassy

desert

coalition

afghanistan

anti−terror

wednesday

capital

taliban

united_states

pres_bush

fire

set

undergrind

store

stockpile

site

salt

oil

million

mexico

gulf

four

deep

crude

coast

cavern

500

barrel

two

support

spokesman

schedule

saudi

response

reporter

price

possible

place

morgan_stanlly_dw

measure

los_angeles

leader

lawmaker

issue

investigation

interview

include

flag

firefighter

early

district

death

crime

controller

chief

car

assault

ashcroft

a.m.

action

war

victim

unprecedented

tragedy

system

south

service

search

rubble

police

percent

passenger

number

news

national

member

mayor_giuliani

limit

life

information

hijacker

ground

florida

fbi

clear

business

boston

big

bank

authority

act

airline

wreckage

west

trade

top

terrorism

tell

team

target

suspect

survivor

state

source

side

send

section

school

say

return

rescuer

rescue

point

personnel

pennsylvania

operation

officer

night

newspaper

new

minute

mayor

market

man

local

late

investigator

intelligence

inc

group

giant

floor

faa

employee

deadly

damage

control

congress

company

collapse

capitol

brief

bin_laden

airliner

aviation

street

huge

history

explosion

evacuation

cause

devastate

emergency

way

traffic

the_worst

smoke

blood

scene

terrorist

television

military

massive

jet

hospital

home

headquarters

federal

crash

agency

airplane

world_trade_ctr

world

worker

work

white_house

washington

twin

tuesday

tower

time

thousand

security

plane

people

pentagon

part

official

office

nation

morning

manhattan

major

low

large

landmark

hour

hijack

government

flight

financial

day

country

commercial

city

center

call

buildngs

buildng

area

american

america

airport

air

aircraft

12

13

14

15

16

17

sep−11

sep−12

sep−16

sep−26

oct−11

terror

attack

new_york

strike

timor

sunday

ship−to−shore

revenge

rehearse

landing

helicopter

east

expectation

casualty

bomb

report

civilian

thursday

stronghold

respond

kandahar

grow

intense

kabul

ultimatum

rule

protester

fresh

embassy

desert

coalition

afghanistan

anti−terror

wednesday

capital

taliban

united_states

pres_bush

fire

set

undergrind

store

stockpile

site

salt

oil

million

mexico

gulf

four

deep

crude

coast

cavern

500

barrel

two

support

spokesman

schedule

saudi

response

reporter

price

possible

place

morgan_stanlly_dw

measure

los_angeles

leader

lawmaker

issue

investigation

interview

include

flag

firefighter

early

district

death

crime

controller

chief

car

assault

ashcroft

a.m.

action

war

victim

unprecedented

tragedy

system

south

service

search

rubble

police

percent

passenger

number

news

national

member

mayor_giuliani

limit

life

information

hijacker

ground

florida

fbi

clear

business

boston

big

bank

authority

act

airline

wreckage

west

trade

top

terrorism

tell

team

target

suspect

survivor

state

source

side

send

section

school

say

return

rescuer

rescue

point

personnel

pennsylvania

operation

officer

night

newspaper

new

minute

mayor

market

man

local

late

investigator

intelligence

inc

group

giant

floor

faa

employee

deadly

damage

control

congress

company

collapse

capitol

brief

bin_laden

airliner

aviation

street

huge

history

explosion

evacuation

cause

devastate

emergency

way

traffic

the_worst

smoke

blood

scene

terrorist

television

military

massive

jet

hospital

home

headquarters

federal

crash

agency

airplane

world_trade_ctr

world

worker

work

white_house

washington

twin

tuesday

tower

time

thousand

security

plane

people

pentagon

part

official

office

nation

morning

manhattan

major

low

large

landmark

hour

hijack

government

flight

financial

day

country

commercial

city

center

call

buildngs

buildng

area

american

america

airport

air

aircraft

12

13

14

15

16

17

Figure 3: Terror news temporal cores at level 12.

Figure 4: Terror news temporal cores at level 6 – dendrogram.

In Figure 5 we present the cluster that contains the main "actors": united_states,washington, taliban, afghanistan, government. The corresponding TQs are dis-played as a heatmap in Figure 6.

The second selected cluster is built around the term anthrax (biologicalwarfare). It is presented in Figure 7 and the corresponding TQs in Figure 8.

12

Page 13: Cores in temporal networks A longitudinal approach

taxiranian

grindmobilization

unarmturkey

pledgeheartlandlogistical

sultanprince

saudi_arabiaextraordinarysix−member

outbreakexclusive

working−classunlicenseddorchester

transferdisease

infectiousdiscovery

haqwalterspringsilver

romneyreed

organizemitt

locatedetrick

fortsupporterhigh−leveldelegation

failurehart

southeastwingsport

straightwaistbandtraditional

daggermobile

columbiaivanwalk

directorhouse

senatechairman

committeewestern

fightembassy

retaliationmeasure

requestvictim

deputydetail

carlevelpoint

spokeswomanstreet

fateradiocamplead

yvonneweeklythrust

sneinehindustrial

detaing7

refusalthreatenpm_blair

uzbekistanethnic

warlordsample

stufflebeemeastern

sportsmanseason

seaboardrecognize

randompotter

minetaloose−knitinspector

dobull

celebrateindianapolis

repairsuspicious

bocaraton

exposureresult

womanindexretailwhite

fallconsumercorporateenvelope

sentimentrespite

publishermichigan

powderchurch

week−longveteran

sayedpolitician

pirpashtun

gailaniblitz

captureupper

storthodoxcathedralnicholas

beachpalmjean

maleckihill

physicianuss

staffernaif

khorumjunioreisold

disruptioncarl

averagebuy

berlusconigarden

johnmission

tom_daschlenbc

scaresacramento

ropehalf

auditoriumflag−waving

smithromandinnercharitybenefit

catholicprobableembark

george_patakispecific

cheerconvention

mazar−i−sharifvajpayee

roundbehari

atalavailability

enlistpartner

referincidence

potentintensifymasood

manufacturerassassinate

kansasboard

brotherking

shahzahir

interioroutskirts

khanaask

khairqanuni

yunisborough

pillarrunoffgreenferrer

fernandodefeatbronx

contentioustakhar

legfront−line

fivearabic

channelfootwall

billowdust

flamesend

smokerepublicsatellite

plantsoviet

tajikistantour

hamiltonvillagebajuarpunish

princetonpatience

malakandmaidan

inconclusiveapprove

detectionweekend

mailroomsrecent

inspectiontrace

approvalrating

preliminarydeliveryenergy

internetinhalation

wallaceramadan

perkinslahore

fdaengland

consulatebradley

clinicdecontamination

statusmarket−sensitive

postingnew_jersey

posthospital

unitdepartment

doctorwest

expertteam

californiawave

haltsouth

vietnamstudentrhetoric

anti−united_stateshotbed

wealthyhide

believecrop−dusting

badgefind

southwestskychefs

metropolitanlsgid

catererdetroit

sabrifrankfurt

riazitaly

russiaasian

fugitivepressure

eyethorn

sukarnoputriproject

indonesianmegawati

populationcease−fire

ivanovguerrilla

harborindonesia

israelpalestinian

allianceenemy

lookorder

europeanleave

speakassistance

chinaseptember

clericsaletony

sweeneysponsor

shortshenzhen

jonashuangtiangulbuddin

dogbattalioncontinue

hekmatyarcabinet−level

creationallege

blairreligious

handspeech

demandwant

frenchedict

guestchildfalse

octcipro

vaccinethompsonproduction

newboldkashmir

icrcalaska

hoaxlaunchprobe

tenzairmail

winterticket

sagmore_and_more

jump−startisa

hamademir

dirksencustomeral−khalifa

bahrainsouth−west

quadrantelevator

freightcorrespondent

foodminority

sulaimanlongtime

bughaith

tanksabon

residentialpoorkano

f/a−18gari

neighborhoodbroadcastal−jazeera

qatarindianindiairan

clausemutual

singhrabbanikuwaitextend

english−languagecounter−terrorism

coordinatoral−fahadal−sabah

tombshebaa

patriarchoutpost

israel−lebanonhebrondivide

farmjack

strawanti−terrorism

anticipationdismiss

apecsplit

strongholdlightning

darkenswab

parachuteford

airfieldclip

sentencenear−realmandate

jailhost

genericbayer

equivalentgiant

killvirginiacausehuge

explosionstrategic

ricecollege

sanrockville

jordaniangrossmont

falmouthdifrancesco

diegobagram

b−52airbaseattendreduce

rabifazilqari

centerhealthpublictroop

campaigntarget

airbomb

bacteriumtestmail

postalanthrax

letterfacilityworkerbuildng

officeauthority

employeeorange

bombersong

anti−aircraftburst

videotapemaireadmaguire

irelandgod

founderaim

corriganship

anticipatedeck

regimehumanitarian

peaceforeigner

remarkdropoffer

vinsonprotest

warplaneprime

aidcoalition

allyforeign

ministerbritish

mohammadmullah

omarkhan

quotebritain

residentisraeli

saturdaypakistani

rulersimilarrichard

secmyers

national_guardgovernor

alannual

fifthrole

entiretomahawk

paradejazeera

columbusclark

ceremonycabinet

agwunobiavenue

radarramstein

thirdpreparedness

half−percentageaggressive

federal_reservecutjohn_mccain

jacquesindiana

evanchirac

arizonabayhsea

politicallawmakerprogram

jointpaul

overnightlibrary

afulashootlower

kennethborrow

katzmanturkish

shopomaghdonaldankara

arieltel

stimulusshevardnadze

rebatepayroll

avivnovosibirsk

releaseadvantageworldwide

morefriendlydetaileddossierprivate

donationgain

anti−talibans

advancelandlocked

aippakistan−based

tomhomeland

ridgeelite

pacificranger

summitjiang

shanghaiinfectionconfirm

skintrentonbrokawphase

chinesenavy

bureaurear

pendnotification

nextkin

fellowadm

deadzemin

remainfemale

cbscombat

associationspreadhuman

supreme_courtdrug

mailroomus_postal_service

torontotask

dara−i−sufsamangan

prisonernotice

forfeitureconstituteadequate

agriculturefilter

off−siteantibiotic

warehouseashcroftmonday

usepossible

reportcity

timegroupleader

daymilitary

strikeweek

countryterrorism

forcegovernmentafghanistan

talibanattack

united_stateswashington

officialpeople

bin_ladenpres_bushpentagonnew_yorkamerican

warnew

securityinternational

nationworld

supportpakistan

actionislamic

manstateeffort

policerule

muslimhijack

world_trade_ctrterrorist

fbiplane

spokesmanaircraft

areareporter

telldefense

threatagency

americarumsfeld

hemember

partwarnfriday

thursdaykandaharsouthern

helicopterspecialsunday

al_quaedamonth

septunited_states−led

afghankabulfuture

capitalflorida

raidministry

hananhimat

provinceuzbek

mastermindaccuseunlikely

possessakhund

obaidullahbank

centralpower

talkadministration

neighborhistory

onlycourt

industrymiddle_east

questioncooperation

kindcongressional

islamabadperpetratormusharraf

aidepervez

risksame

attorneyhighwayfinance

initialabu

hardlineinstitute

red_crossadviser

rebelyouth

turkmenistantrack

tinystrikingsolana

secrecysayyaf

prizepresence

otmarnorwegian

nominationnobel

niyazovnigel

mubarakmissouri

mcfaulmaintainlightfoot

kyrgyzstanislamistinterpol

hesitancyhelmick

haslergephardt

experiencedwightalpine

distancediplomatic

strictpropose

michelhear

deploymentaffair

chechenbill

milliondemocratic

patrolinvolvement

japanesemake

leadershiphouse_of_reps

righttie

moscowspike

kazakhstanpotential

oceanconfidence

conductsurveyability

parenthoodfrank

federationdepts

abortionday−old

winwidespread

around−the−clockpraise

mosquecover

scurrysaudi_arabian

realitymecca

grimcommonwealth

grandcite

holyabdullah

worshipperal−nahayan

zaidbin

sheikhaustraliahijacking

visitsign

travelwarning

lawyernevada

litigationbrazil

jacobabaddostum

rashidabdul

commanderbioterrorism

germwarfare

evidenceradical

germanycrowdpass

relationunion

proposallaboratoryresolution

intensefaction

ulema−i−islamsinclairshatter

pro−talibanpredicament

muraijin

jamiathamburg

commissionactivation

brkarachi

gearcommerceabc_news

closurejfk

los_angeles−boundafternoon

flagmighty

suspendjohn_f_kennedy

kennedymohamed

famousauthorize

rainyahya

tunisiapull

moroccohabib

electronicdwindle

doverdelaware

benbenaissa

cockpitprosecutorbipartisan

constructioncustodyapplaud

motorcadenew_yorkers

remembrancemight

attackerferry

air_forcememorial

abcoil

boundpaper

reservesenator

protectiondrive

rescuerdevastation

arreststandhotelview

extremistisolate

capacityprayer

disasterrecovery

justice_deptbackport

researchmaryland

vladimirtracy

shi'itekharrazi

kamalballingerbahram

azizahad

al−rahmanqueen

a300airbus

substanceerin

abuseartigianiahmad

putinrussian

natoword

aformer_pres_clinton

islamultimatum

sessiontrouble

sermonseizure

sanctionreference

magistratejudge

komiveshard−hitting

negativeN1000

fatwacanadian

policybroadbattle

immigrationgulf

decisioneu

assetphotograph

teeny−bopperssqueaky

roilopresidential

philippinemary−kate

legendgolez

freezecomicclean

98−year−oldbob

sergeidiscussion

fringetouristdisney

waltfresh

protesteranti−terror

desertsurveillance

thomasunited_arab_enerates

assistantwal

vallieresikh

schwabretire

oklahomanickle

lieutenantlecturer

gregdon

daawaconflict

cia_dir_tenetcboe

burnettealgerian

bailshelby

90−percentenjoy

mohammedal−maktoum

jihad8

legislationfrance

givecriminal

proofaccountattempt

assassinationmurder

parisnegotiation

liberalkitchener

fund−raisergang

staggerfashionbackup

7:007:30

shadowyskyscraper

worryworld_bank

whitmanvergetrader

speculationslump

severeresume

restrictionraft

promptoracle

onemica

marwanjarvis

inimf

gapedenydeal

crowleycotton

complexchar

centurycash

cancelbullionbeijing

armitageadvertiseal−shehhistringent

stillreceiveqaeda

patternorigin

nymexnetegrity

imposehaydenextract

econversation

creditusa

significantlatin

deskfimat

globereel

texasaerial

identificationdeclare

engineerso−called

commandeertwist

equipmentkerikruin

muellerlimit

wreckagejetliner

recorderfinal

remainsneed

recessiondissident

banweb

issueuae

possibilitytransportation

dataton

bodyfuel

interestrate

focusassociate

clearmorgan_stanlly_dw

containdeutsche

equitybox

figuremarknybot

passportentertainment

dollarafrica

divisioninvestment

bagbond

commoditycondition

claimfund

airspaceblackwaterutility

sunphoto

infrastructuredrinking

editor63−year−old

tabloidcard

estimatefriend

processcriticism

handleopponent

three−week−oldmartyrdomgruesomeexecution

6dear

townshipmedication

handlerfraction

capitol−areadistribution

deltavilasrao

maharashtradeshmukh

bombaycommando

spotsocialist

shoreportrayal

disparage28th

berkeleytheodore

f−14roosevelt

undergrindstockpile

saltcrude

cavern500

barrelcontroller

crimeterminal

departuremassachusetts

tenantresumptionrestaurant

refineryportland

picturenoon

motorolamanager

logankorea

jerusalemisland

investorhorrendous

gmcorp

chancearrangement

attentionstructure

riverrental

regularreasonportionlicense

extraherald

strategyresource

reviewag

situationdramatic

filmpowerful

confirmationmaker

alertyesterday

yenwide

wholesalewatch

warnervirtual

vengeanceunnamed

unknowtravelerterrible

tamillowstory

storagespend

sonysomerset

smoldersingapore

sarasotarequirement

providepremature.afghanistan

platformpentagon.the

patriotismpage

operatorongoing

northeastnablus

mikelot

livelistener

lineuplee

koekiknife−wielding

kintonjunejump

inter−servicesillinois

hoglanhigh.thehammer

hguy

grayfreedom−loving

flagshipfichtelfairfax

extraditionextradite

exportease

culpritcredible

coveragecorridorcornerconed

comedyco−founder

charlesbroker

briefingblacken

beawait

aolannouncement

americaresamerican_airline

amazonabdul_salam_zaeef

acewtc

wsvnveronique

veniceuniform

tragictape

supplierskyline

shesecretary−general

saddamrun

robertsonprecautionary

plaugherpayment

pathmovie

michaelmassport

maldonadolufthansakamikaze

joint_chiefs_ofstaffinstrument

injurehollywood

good_morning_americaframe

enterprisedirection

digdifferent

desperatedaughter

contentcollenettecolleaguechitwood

caseycairn

bellevueamtrakairway

adult619

28−year−old52−year

waitvolunteer

vancouverundetermined

tightstrand

sinopecs−oil

revenuerecover

producerpetroleum

norman_minetamotoristmothermanualmaine

ltdlose

logan_intl_airporthorrible

greenspangasoline

divertdec

culturecontact

coandrew

42ndabrahamreaction

startcitizen

takesecond

heightenreprisalexpress

earlypackage

economiceconomy

globaljob

heavyroom

addressplan

cuteurope

ambassadorimpact

stafftoll

districtmainplaceiraqi

moveinterview

newspaperhope

wall_streetprice

hittraffic

iraqsupply

deepdelay

shutdownunidentified

majorityfox

innocentpositionborder

exileshow

percentageoctober

unemploymentshot4.95.4

dealerconfusion

24alarmwarm

regionalpositive

leducpolicy−setting

daviscdc

preventionbay

goldenextent

primarygate

bridgesan_francisco

chargeregionjustice

universitygerman

stepfour

govoice

precautionsimultaneous

witnesscivil

studyclose

stationbrief

secretarycruise

missilehigh

messagenorthern

fighternuclear

armyrepublican

activitybrussels

analystwolfowitz

riserose

countyrevenge

checksectiontragedy

piecetransport

miamitelecommunication

relieftechnology

tuckerevening

sensefiremanwarrant

aftermathmountainproblem

standarddoom

ari_fleischerconnection

lowresponsibility

townreal

strongdemocrat

japangood

individualmedical

milecrush

holesingle

reopentechnical

armconcrete

ruraladd

meetingfield

normalspanish

mariaaznar

josesecluderetreat

putplain−clothes

midtownmarshal

kjavitsjacobfavor

2−1/2catoctinwireless

signalbochum

crippleunanimous

santorumprocedure

materialmagazine

itemguided

glass−walledemotional

cruisercommunist

changecall−up

byeraegisallow

solidarityseparateschumerreservist

readinesspreparation

phonyformal

enormouselectrical

authorizationdeclaration

moneyform

commissionerthingparty

optionrefugee

pollguardopenseekcold

nasdaqnyse

accessgrasso

greatpress

struggledestroyliability

insurancelondon

collapselos_angeles

carryschool

steelmanagement

metala.m.p.m

knifespace

communitywe

unprecedentedfull

the_worstcosteast

coastmassive

cnnlebanon

noimmediate

previousdestruction

massshare

soilhelpown

basiscrisis

footagejewish

bushold

sharonvincent

veryutter

tremendoustreatment

territoryterrify

suburbanstream

stoxxstop

sternsteadysorrow

sistersilent

sidewalkserioussector

sacresolve

rejectionreinsurer

rawpedestrian

patientparalyze

oklahoma_cityoffutt

ny1nasa

middlememory

laguardiakenyans

jonesjerryjanehunt

hartwiggun

gridlockglass

freefoundation

flowengine

electioneight−month

dowdonor

d.ccondolencecompound

colombiacloud

clintonchris

chaoscandle

busloadblast

barksdalebacker

appropriateambulance

199839−year−old

yankeeworld_war_2

themestretcher

stretchstadium

squiresophisticate

sirenshoerushreek

railoutsideorlando

oldnetanyahu

mexicanlouisiana

lincolngame

fleecustom

crumbleco−ordination

chaincerrone

cemeterybiz−rberlin

available39−year

acridinsurer

untellyates

sortproperty

mournmodern

mallmainlandextensive

developmentdestination

communicationapparentbombing

rednebraska

eyewitnesscross

evacuatesoldiertough

commandflash

controlairplane

trydept

heartseries

floorreturn

endsubway

directincident

commentyasser_arafat

increasefly

northnearbyshake

tripcasualtyrespond

bloodcommuter

vehicledowntown

canadaolsonblow

evacuationhorrorinjurynerve

pearl_barborhusband

tallbad

dullesburn

shockwindow

tedsolicitor

getnewark

woodpanic

carnageimageyoung

presidencyhorrific

lightsafe

momentarlingtonlocation

pittsburghsecret

traumastore

georgewife

variousvowgmt

potomacedt

corporationsears_tower

mexicosetcia

sweepviolation

amirmuttaqi

1991pact

inhaleenvironmental

educationagreement

breachnichols

danlt

hizbollahtimor

ship−to−shorelanding

rehearseshelterweigh

al−midharurgent

war−erasyria

sharpreagan

prospectpier

navalmultimillionaire

mountmind

manilamanhunt

maneuverinside

identifyhard

exercisedelhi

decline19

capabilitysympathy

pres_bushialhatred

practiceroutineblame

expandchemical

expectationhard−linebiological

weaponface

spiritualsupremejournalist

ridleyarabic−language

sheltonpuritanical

okazex−king

atifarrival

189ali

documentzerou.n

councilurgegrow

jamespanel

judiciaryleahy

koizumidiplomat

importantopinion

australiancallan

30agricultural

zhuwam

religionrahmanpatrickmentor

iranian−backedhidehiko

gatherfollower

delraycomputercallahanbangzao

bangladesh130

anti−terroristschieffer

satojakartafourth

howardvp_cheney

two−dayachakzai

nematullahanchorpound

nomguerre

deatty_gen_ashcroftsaid

12thatef

egyptiananthrax−laced

fierceairporthome

siteagent

systemconference

newstop

enforcementlaw

theywakebase

suicidewednesday

terrortuesday

informationsuspect

yearorganization

networkmilitant

saudi−bornchieflarge

officerfinancial

airlinehijacker

misspennsylvania

nationalnumberpercent

responsereuter

keysay

congresssourceformer

movementgeneral

namesenior

statementmeet

presidentarabway

majorintelligence

white_housecarrier

firefederal

workinclude

operationgroundcapitolsaudi

concernlate

casecivilianserviceairliner

thousandcall

headquarterstradelocal

mayormayor_giuliani

fearnight

mediaperson

itfamily

twoinvestigation

biglink

deadlyflight

passengertower

twinjetlist

televisionsmall

assaulthour

morningbuildngs

companyfirst

eventbusiness

personnelinvestigatorcoordinate

united_airlinesamerican_airlines

ofdeathscene

damagesymbolboston

totalminute

faasafety

manhattanside

commerciallife

pilotpowell

actsurvivor

debris110−story

followvicefirmtrain

aviationloss

telephonelong

blockschedule

phoneboeing

celldozen

defense_deptsearch

hundredrescuerubble

emergencydevastate

headchicago

crewfirefighterlandmark

marketexchange

stockcrash

domesticannounce

parkarabianunusual

threesergeant

peninsulapecker

outcomehigh−rise

hangexpatriate

e−tradecontamination

brokeragebattery

asbestosapartment

10−year−oldami

davidson

onlinerelative

incexecutive

skytreasurycurrent

doubtactive

dutyshift

versionsensenbrenner

safariniplot

millenniumliberty

explosivechallenge

cast1.4

biodefensequadrennial

three−monthbrentwood

involvemortalitymaternal

117

frontline11

northern_allianceopposition

0.0 0.4 0.8

Terror cores 6/Com

plete

hclust (*, "complete")

D

Height

Figure 5: Terror news temporal cores at level 6 – cluster 1.

*** Comment senator, date; entered laterThe described approach can be used on up to some thousands of nodes.

Another approach to large sets of nodes is to consider only the most activenodes. The activity of the TQ a can be measured by its aggregated value [5]

Σa =∑

i

(fi − si) · vi

The procedure is now simple – order the nodes in decreasing order with respectto their activity and extract the g ∈ N top one. Display them with a heatmap.For terror news network cores, g = 30, it is presented in Figure 9.

*** but time span of these cores is wider, not limited mostly to first 11 days.The max pS-core numbers through time interval is shown in Fig. ??. It drops

13

Page 14: Cores in temporal networks A longitudinal approach

sep−11

sep−12

sep−13

sep−14

sep−15

sep−16

sep−17

sep−18

sep−19

sep−20

sep−21

sep−22

sep−24

sep−25

sep−26

sep−27

sep−28

sep−29

sep−30

oct−1

oct−2

oct−3

oct−4

oct−5

oct−6

oct−7

oct−8

oct−10

oct−11

oct−12

oct−13

oct−14

oct−15

oct−16

oct−17

oct−18

oct−19

oct−20

oct−21

oct−22

oct−23

oct−24

oct−25

oct−26

oct−27

oct−28

oct−29

oct−30

oct−31

nov−1

nov−2

nov−3

nov−4

nov−5

nov−6

nov−7

nov−8

nov−9

nov−10

nov−11

nov−12

nov−13

nov−14

nov−15

poss

ible

repo

rt

city

time

grou

p

lead

er

day

mili

tary

strik

e

wee

k

coun

try

terr

oris

m

forc

e

gove

rnm

ent

afgh

anis

tan

talib

an

atta

ck

unite

d_st

ates

was

hing

ton

offic

ial

peop

le

bin_

lade

n

pres

_bus

h

pent

agon

new

_yor

k

amer

ican

war

new

secu

rity

inte

rnat

iona

l

natio

n

wor

ld

supp

ort

paki

stan

actio

n

isla

mic

man

stat

e

effo

rt

polic

e

rule

mus

lim

hija

ck

wor

ld_t

rade

_ctr

terr

oris

t

fbi

plan

e

6810121416

Figu

re6:

Terror

newstempo

ralc

ores

atlevel6

–cluster1.

14

Page 15: Cores in temporal networks A longitudinal approach

taxiranian

grindmobilization

unarmturkey

pledgeheartlandlogistical

sultanprince

saudi_arabiaextraordinarysix−member

outbreakexclusive

working−classunlicenseddorchester

transferdisease

infectiousdiscovery

haqwalterspringsilver

romneyreed

organizemitt

locatedetrick

fortsupporterhigh−leveldelegation

failurehart

southeastwingsport

straightwaistbandtraditional

daggermobile

columbiaivanwalk

directorhouse

senatechairman

committeewestern

fightembassy

retaliationmeasure

requestvictim

deputydetail

carlevelpoint

spokeswomanstreet

fateradiocamplead

yvonneweeklythrust

sneinehindustrial

detaing7

refusalthreatenpm_blair

uzbekistanethnic

warlordsample

stufflebeemeastern

sportsmanseason

seaboardrecognize

randompotter

minetaloose−knitinspector

dobull

celebrateindianapolis

repairsuspicious

bocaraton

exposureresult

womanindexretailwhite

fallconsumercorporateenvelope

sentimentrespite

publishermichigan

powderchurch

week−longveteran

sayedpolitician

pirpashtun

gailaniblitz

captureupper

storthodoxcathedralnicholas

beachpalmjean

maleckihill

physicianuss

staffernaif

khorumjunioreisold

disruptioncarl

averagebuy

berlusconigarden

johnmission

tom_daschlenbc

scaresacramento

ropehalf

auditoriumflag−waving

smithromandinnercharitybenefit

catholicprobableembark

george_patakispecific

cheerconvention

mazar−i−sharifvajpayee

roundbehari

atalavailability

enlistpartner

referincidence

potentintensifymasood

manufacturerassassinate

kansasboard

brotherking

shahzahir

interioroutskirts

khanaask

khairqanuni

yunisborough

pillarrunoffgreenferrer

fernandodefeatbronx

contentioustakhar

legfront−line

fivearabic

channelfootwall

billowdust

flamesend

smokerepublicsatellite

plantsoviet

tajikistantour

hamiltonvillagebajuarpunish

princetonpatience

malakandmaidan

inconclusiveapprove

detectionweekend

mailroomsrecent

inspectiontrace

approvalrating

preliminarydeliveryenergy

internetinhalation

wallaceramadan

perkinslahore

fdaengland

consulatebradley

clinicdecontamination

statusmarket−sensitive

postingnew_jersey

posthospital

unitdepartment

doctorwest

expertteam

californiawave

haltsouth

vietnamstudentrhetoric

anti−united_stateshotbed

wealthyhide

believecrop−dusting

badgefind

southwestskychefs

metropolitanlsgid

catererdetroit

sabrifrankfurt

riazitaly

russiaasian

fugitivepressure

eyethorn

sukarnoputriproject

indonesianmegawati

populationcease−fire

ivanovguerrilla

harborindonesia

israelpalestinian

allianceenemy

lookorder

europeanleave

speakassistance

chinaseptember

clericsaletony

sweeneysponsor

shortshenzhen

jonashuangtiangulbuddin

dogbattalioncontinue

hekmatyarcabinet−level

creationallege

blairreligious

handspeech

demandwant

frenchedict

guestchildfalse

octcipro

vaccinethompsonproduction

newboldkashmir

icrcalaska

hoaxlaunchprobe

tenzairmail

winterticket

sagmore_and_more

jump−startisa

hamademir

dirksencustomeral−khalifa

bahrainsouth−west

quadrantelevator

freightcorrespondent

foodminority

sulaimanlongtime

bughaith

tanksabon

residentialpoorkano

f/a−18gari

neighborhoodbroadcastal−jazeera

qatarindianindiairan

clausemutual

singhrabbanikuwaitextend

english−languagecounter−terrorism

coordinatoral−fahadal−sabah

tombshebaa

patriarchoutpost

israel−lebanonhebrondivide

farmjack

strawanti−terrorism

anticipationdismiss

apecsplit

strongholdlightning

darkenswab

parachuteford

airfieldclip

sentencenear−realmandate

jailhost

genericbayer

equivalentgiant

killvirginiacausehuge

explosionstrategic

ricecollege

sanrockville

jordaniangrossmont

falmouthdifrancesco

diegobagram

b−52airbaseattendreduce

rabifazilqari

centerhealthpublictroop

campaigntarget

airbomb

bacteriumtestmail

postalanthrax

letterfacilityworkerbuildng

officeauthority

employeeorange

bombersong

anti−aircraftburst

videotapemaireadmaguire

irelandgod

founderaim

corriganship

anticipatedeck

regimehumanitarian

peaceforeigner

remarkdropoffer

vinsonprotest

warplaneprime

aidcoalition

allyforeign

ministerbritish

mohammadmullah

omarkhan

quotebritain

residentisraeli

saturdaypakistani

rulersimilarrichard

secmyers

national_guardgovernor

alannual

fifthrole

entiretomahawk

paradejazeera

columbusclark

ceremonycabinet

agwunobiavenue

radarramstein

thirdpreparedness

half−percentageaggressive

federal_reservecutjohn_mccain

jacquesindiana

evanchirac

arizonabayhsea

politicallawmakerprogram

jointpaul

overnightlibrary

afulashootlower

kennethborrow

katzmanturkish

shopomaghdonaldankara

arieltel

stimulusshevardnadze

rebatepayroll

avivnovosibirsk

releaseadvantageworldwide

morefriendlydetaileddossierprivate

donationgain

anti−talibans

advancelandlocked

aippakistan−based

tomhomeland

ridgeelite

pacificranger

summitjiang

shanghaiinfectionconfirm

skintrentonbrokawphase

chinesenavy

bureaurear

pendnotification

nextkin

fellowadm

deadzemin

remainfemale

cbscombat

associationspreadhuman

supreme_courtdrug

mailroomus_postal_service

torontotask

dara−i−sufsamangan

prisonernotice

forfeitureconstituteadequate

agriculturefilter

off−siteantibiotic

warehouseashcroftmonday

usepossible

reportcity

timegroupleader

daymilitary

strikeweek

countryterrorism

forcegovernmentafghanistan

talibanattack

united_stateswashington

officialpeople

bin_ladenpres_bushpentagonnew_yorkamerican

warnew

securityinternational

nationworld

supportpakistan

actionislamic

manstateeffort

policerule

muslimhijack

world_trade_ctrterrorist

fbiplane

spokesmanaircraft

areareporter

telldefense

threatagency

americarumsfeld

hemember

partwarnfriday

thursdaykandaharsouthern

helicopterspecialsunday

al_quaedamonth

septunited_states−led

afghankabulfuture

capitalflorida

raidministry

hananhimat

provinceuzbek

mastermindaccuseunlikely

possessakhund

obaidullahbank

centralpower

talkadministration

neighborhistory

onlycourt

industrymiddle_east

questioncooperation

kindcongressional

islamabadperpetratormusharraf

aidepervez

risksame

attorneyhighwayfinance

initialabu

hardlineinstitute

red_crossadviser

rebelyouth

turkmenistantrack

tinystrikingsolana

secrecysayyaf

prizepresence

otmarnorwegian

nominationnobel

niyazovnigel

mubarakmissouri

mcfaulmaintainlightfoot

kyrgyzstanislamistinterpol

hesitancyhelmick

haslergephardt

experiencedwightalpine

distancediplomatic

strictpropose

michelhear

deploymentaffair

chechenbill

milliondemocratic

patrolinvolvement

japanesemake

leadershiphouse_of_reps

righttie

moscowspike

kazakhstanpotential

oceanconfidence

conductsurveyability

parenthoodfrank

federationdepts

abortionday−old

winwidespread

around−the−clockpraise

mosquecover

scurrysaudi_arabian

realitymecca

grimcommonwealth

grandcite

holyabdullah

worshipperal−nahayan

zaidbin

sheikhaustraliahijacking

visitsign

travelwarning

lawyernevada

litigationbrazil

jacobabaddostum

rashidabdul

commanderbioterrorism

germwarfare

evidenceradical

germanycrowdpass

relationunion

proposallaboratoryresolution

intensefaction

ulema−i−islamsinclairshatter

pro−talibanpredicament

muraijin

jamiathamburg

commissionactivation

brkarachi

gearcommerceabc_news

closurejfk

los_angeles−boundafternoon

flagmighty

suspendjohn_f_kennedy

kennedymohamed

famousauthorize

rainyahya

tunisiapull

moroccohabib

electronicdwindle

doverdelaware

benbenaissa

cockpitprosecutorbipartisan

constructioncustodyapplaud

motorcadenew_yorkers

remembrancemight

attackerferry

air_forcememorial

abcoil

boundpaper

reservesenator

protectiondrive

rescuerdevastation

arreststandhotelview

extremistisolate

capacityprayer

disasterrecovery

justice_deptbackport

researchmaryland

vladimirtracy

shi'itekharrazi

kamalballingerbahram

azizahad

al−rahmanqueen

a300airbus

substanceerin

abuseartigianiahmad

putinrussian

natoword

aformer_pres_clinton

islamultimatum

sessiontrouble

sermonseizure

sanctionreference

magistratejudge

komiveshard−hitting

negativeN1000

fatwacanadian

policybroadbattle

immigrationgulf

decisioneu

assetphotograph

teeny−bopperssqueaky

roilopresidential

philippinemary−kate

legendgolez

freezecomicclean

98−year−oldbob

sergeidiscussion

fringetouristdisney

waltfresh

protesteranti−terror

desertsurveillance

thomasunited_arab_enerates

assistantwal

vallieresikh

schwabretire

oklahomanickle

lieutenantlecturer

gregdon

daawaconflict

cia_dir_tenetcboe

burnettealgerian

bailshelby

90−percentenjoy

mohammedal−maktoum

jihad8

legislationfrance

givecriminal

proofaccountattempt

assassinationmurder

parisnegotiation

liberalkitchener

fund−raisergang

staggerfashionbackup

7:007:30

shadowyskyscraper

worryworld_bank

whitmanvergetrader

speculationslump

severeresume

restrictionraft

promptoracle

onemica

marwanjarvis

inimf

gapedenydeal

crowleycotton

complexchar

centurycash

cancelbullionbeijing

armitageadvertiseal−shehhistringent

stillreceiveqaeda

patternorigin

nymexnetegrity

imposehaydenextract

econversation

creditusa

significantlatin

deskfimat

globereel

texasaerial

identificationdeclare

engineerso−called

commandeertwist

equipmentkerikruin

muellerlimit

wreckagejetliner

recorderfinal

remainsneed

recessiondissident

banweb

issueuae

possibilitytransportation

dataton

bodyfuel

interestrate

focusassociate

clearmorgan_stanlly_dw

containdeutsche

equitybox

figuremarknybot

passportentertainment

dollarafrica

divisioninvestment

bagbond

commoditycondition

claimfund

airspaceblackwaterutility

sunphoto

infrastructuredrinking

editor63−year−old

tabloidcard

estimatefriend

processcriticism

handleopponent

three−week−oldmartyrdomgruesomeexecution

6dear

townshipmedication

handlerfraction

capitol−areadistribution

deltavilasrao

maharashtradeshmukh

bombaycommando

spotsocialist

shoreportrayal

disparage28th

berkeleytheodore

f−14roosevelt

undergrindstockpile

saltcrude

cavern500

barrelcontroller

crimeterminal

departuremassachusetts

tenantresumptionrestaurant

refineryportland

picturenoon

motorolamanager

logankorea

jerusalemisland

investorhorrendous

gmcorp

chancearrangement

attentionstructure

riverrental

regularreasonportionlicense

extraherald

strategyresource

reviewag

situationdramatic

filmpowerful

confirmationmaker

alertyesterday

yenwide

wholesalewatch

warnervirtual

vengeanceunnamed

unknowtravelerterrible

tamillowstory

storagespend

sonysomerset

smoldersingapore

sarasotarequirement

providepremature.afghanistan

platformpentagon.the

patriotismpage

operatorongoing

northeastnablus

mikelot

livelistener

lineuplee

koekiknife−wielding

kintonjunejump

inter−servicesillinois

hoglanhigh.thehammer

hguy

grayfreedom−loving

flagshipfichtelfairfax

extraditionextradite

exportease

culpritcredible

coveragecorridorcornerconed

comedyco−founder

charlesbroker

briefingblacken

beawait

aolannouncement

americaresamerican_airline

amazonabdul_salam_zaeef

acewtc

wsvnveronique

veniceuniform

tragictape

supplierskyline

shesecretary−general

saddamrun

robertsonprecautionary

plaugherpayment

pathmovie

michaelmassport

maldonadolufthansakamikaze

joint_chiefs_ofstaffinstrument

injurehollywood

good_morning_americaframe

enterprisedirection

digdifferent

desperatedaughter

contentcollenettecolleaguechitwood

caseycairn

bellevueamtrakairway

adult619

28−year−old52−year

waitvolunteer

vancouverundetermined

tightstrand

sinopecs−oil

revenuerecover

producerpetroleum

norman_minetamotoristmothermanualmaine

ltdlose

logan_intl_airporthorrible

greenspangasoline

divertdec

culturecontact

coandrew

42ndabrahamreaction

startcitizen

takesecond

heightenreprisalexpress

earlypackage

economiceconomy

globaljob

heavyroom

addressplan

cuteurope

ambassadorimpact

stafftoll

districtmainplaceiraqi

moveinterview

newspaperhope

wall_streetprice

hittraffic

iraqsupply

deepdelay

shutdownunidentified

majorityfox

innocentpositionborder

exileshow

percentageoctober

unemploymentshot4.95.4

dealerconfusion

24alarmwarm

regionalpositive

leducpolicy−setting

daviscdc

preventionbay

goldenextent

primarygate

bridgesan_francisco

chargeregionjustice

universitygerman

stepfour

govoice

precautionsimultaneous

witnesscivil

studyclose

stationbrief

secretarycruise

missilehigh

messagenorthern

fighternuclear

armyrepublican

activitybrussels

analystwolfowitz

riserose

countyrevenge

checksectiontragedy

piecetransport

miamitelecommunication

relieftechnology

tuckerevening

sensefiremanwarrant

aftermathmountainproblem

standarddoom

ari_fleischerconnection

lowresponsibility

townreal

strongdemocrat

japangood

individualmedical

milecrush

holesingle

reopentechnical

armconcrete

ruraladd

meetingfield

normalspanish

mariaaznar

josesecluderetreat

putplain−clothes

midtownmarshal

kjavitsjacobfavor

2−1/2catoctinwireless

signalbochum

crippleunanimous

santorumprocedure

materialmagazine

itemguided

glass−walledemotional

cruisercommunist

changecall−up

byeraegisallow

solidarityseparateschumerreservist

readinesspreparation

phonyformal

enormouselectrical

authorizationdeclaration

moneyform

commissionerthingparty

optionrefugee

pollguardopenseekcold

nasdaqnyse

accessgrasso

greatpress

struggledestroyliability

insurancelondon

collapselos_angeles

carryschool

steelmanagement

metala.m.p.m

knifespace

communitywe

unprecedentedfull

the_worstcosteast

coastmassive

cnnlebanon

noimmediate

previousdestruction

massshare

soilhelpown

basiscrisis

footagejewish

bushold

sharonvincent

veryutter

tremendoustreatment

territoryterrify

suburbanstream

stoxxstop

sternsteadysorrow

sistersilent

sidewalkserioussector

sacresolve

rejectionreinsurer

rawpedestrian

patientparalyze

oklahoma_cityoffutt

ny1nasa

middlememory

laguardiakenyans

jonesjerryjanehunt

hartwiggun

gridlockglass

freefoundation

flowengine

electioneight−month

dowdonor

d.ccondolencecompound

colombiacloud

clintonchris

chaoscandle

busloadblast

barksdalebacker

appropriateambulance

199839−year−old

yankeeworld_war_2

themestretcher

stretchstadium

squiresophisticate

sirenshoerushreek

railoutsideorlando

oldnetanyahu

mexicanlouisiana

lincolngame

fleecustom

crumbleco−ordination

chaincerrone

cemeterybiz−rberlin

available39−year

acridinsurer

untellyates

sortproperty

mournmodern

mallmainlandextensive

developmentdestination

communicationapparentbombing

rednebraska

eyewitnesscross

evacuatesoldiertough

commandflash

controlairplane

trydept

heartseries

floorreturn

endsubway

directincident

commentyasser_arafat

increasefly

northnearbyshake

tripcasualtyrespond

bloodcommuter

vehicledowntown

canadaolsonblow

evacuationhorrorinjurynerve

pearl_barborhusband

tallbad

dullesburn

shockwindow

tedsolicitor

getnewark

woodpanic

carnageimageyoung

presidencyhorrific

lightsafe

momentarlingtonlocation

pittsburghsecret

traumastore

georgewife

variousvowgmt

potomacedt

corporationsears_tower

mexicosetcia

sweepviolation

amirmuttaqi

1991pact

inhaleenvironmental

educationagreement

breachnichols

danlt

hizbollahtimor

ship−to−shorelanding

rehearseshelterweigh

al−midharurgent

war−erasyria

sharpreagan

prospectpier

navalmultimillionaire

mountmind

manilamanhunt

maneuverinside

identifyhard

exercisedelhi

decline19

capabilitysympathy

pres_bushialhatred

practiceroutineblame

expandchemical

expectationhard−linebiological

weaponface

spiritualsupremejournalist

ridleyarabic−language

sheltonpuritanical

okazex−king

atifarrival

189ali

documentzerou.n

councilurgegrow

jamespanel

judiciaryleahy

koizumidiplomat

importantopinion

australiancallan

30agricultural

zhuwam

religionrahmanpatrickmentor

iranian−backedhidehiko

gatherfollower

delraycomputercallahanbangzao

bangladesh130

anti−terroristschieffer

satojakartafourth

howardvp_cheney

two−dayachakzai

nematullahanchorpound

nomguerre

deatty_gen_ashcroftsaid

12thatef

egyptiananthrax−laced

fierceairporthome

siteagent

systemconference

newstop

enforcementlaw

theywakebase

suicidewednesday

terrortuesday

informationsuspect

yearorganization

networkmilitant

saudi−bornchieflarge

officerfinancial

airlinehijacker

misspennsylvania

nationalnumberpercent

responsereuter

keysay

congresssourceformer

movementgeneral

namesenior

statementmeet

presidentarabway

majorintelligence

white_housecarrier

firefederal

workinclude

operationgroundcapitolsaudi

concernlate

casecivilianserviceairliner

thousandcall

headquarterstradelocal

mayormayor_giuliani

fearnight

mediaperson

itfamily

twoinvestigation

biglink

deadlyflight

passengertower

twinjetlist

televisionsmall

assaulthour

morningbuildngs

companyfirst

eventbusiness

personnelinvestigatorcoordinate

united_airlinesamerican_airlines

ofdeathscene

damagesymbolboston

totalminute

faasafety

manhattanside

commerciallife

pilotpowell

actsurvivor

debris110−story

followvicefirmtrain

aviationloss

telephonelong

blockschedule

phoneboeing

celldozen

defense_deptsearch

hundredrescuerubble

emergencydevastate

headchicago

crewfirefighterlandmark

marketexchange

stockcrash

domesticannounce

parkarabianunusual

threesergeant

peninsulapecker

outcomehigh−rise

hangexpatriate

e−tradecontamination

brokeragebattery

asbestosapartment

10−year−oldami

davidson

onlinerelative

incexecutive

skytreasurycurrent

doubtactive

dutyshift

versionsensenbrenner

safariniplot

millenniumliberty

explosivechallenge

cast1.4

biodefensequadrennial

three−monthbrentwood

involvemortalitymaternal

117

frontline11

northern_allianceopposition

0.0 0.4 0.8

Terror cores 6/Com

plete

hclust (*, "complete")

D

Height

Figure 7: Terror news temporal cores at level 6 – cluster 2.

sep−11

sep−12

sep−13

sep−14

sep−15

sep−16

sep−17

sep−18

sep−19

sep−20

sep−21

sep−22

sep−24

sep−25

sep−26

sep−27

sep−28

sep−30

oct−1

oct−2

oct−3

oct−4

oct−5

oct−6

oct−7

oct−8

oct−9

oct−10

oct−11

oct−12

oct−13

oct−14

oct−15

oct−16

oct−17

oct−18

oct−19

oct−20

oct−21

oct−22

oct−23

oct−24

oct−25

oct−26

oct−27

oct−28

oct−29

oct−30

oct−31

nov−1

nov−2

nov−3

nov−4

nov−5

nov−6

nov−7

nov−8

nov−9

nov−12

nov−13

nov−14

nov−15

center

health

public

troop

campaign

target

air

bomb

bacterium

test

mail

postal

anthrax

letter

facility

worker

buildng

office

authority

employee

6

8

10

12

14

Figure 8: Terror news temporal cores at level 6 – cluster 2.

sep−11

sep−12

sep−13

sep−14

sep−15

sep−16

sep−17

sep−18

sep−19

sep−20

sep−21

sep−22

sep−23

sep−24

sep−25

sep−26

sep−27

sep−28

sep−29

sep−30

oct−1

oct−2

oct−3

oct−4

oct−5

oct−6

oct−7

oct−8

oct−9

oct−10

oct−11

oct−12

oct−13

oct−14

oct−15

oct−16

oct−17

oct−18

oct−19

oct−20

oct−21

oct−22

oct−23

oct−24

oct−25

oct−26

oct−27

oct−28

oct−29

oct−30

oct−31

nov−1

nov−2

nov−3

nov−4

nov−5

nov−6

nov−7

nov−8

nov−9

nov−10

nov−11

nov−12

nov−13

nov−14

nov−15

united_states

afghanistan

taliban

attack

official

pres_bush

bin_laden

people

washington

country

military

government

war

new_york

american

day

force

city

leader

strike

week

terrorism

security

time

tell

air

group

new

pentagon

world

5

10

15

Figure 9: Terror news temporal cores at all levels – most active.

rapidly after six days which is also consistent with the idea of news getting oldwith passing time.

4.2 Authors from the field of social network analysis till2018

As an example of application of temporal weighted degree cores we will analyzethe cummulative fractional collaboration (co-authorship) network obtained fromthe collection of networks on the topic of social network analyzed in the papers

15

Page 16: Cores in temporal networks A longitudinal approach

Table 3: Authors with the largest accumulated fractional valueTABELA

[19]. The networks were constructed from the Web of Science data. For detailssee the paper [19].

We start with the two-mode network WA (works× authors) and the vector ywhich contains for each work (paper, book, report) its publication year. Thereare 70792 works and 93011 authors. We combine them into a cummulativetemporal network WAtc = [watc(w, a)], where watc(w, a) = [(y(w), Tmax, 1)]for (w, a) ∈ LWA [8]. The TQ [(y(w), Tmax, 1)], the weight of the link (w, a),essentialy says that this link from the work w to its author a is active from theyear of publication of the work w on.

We extend the standard normalization (fractional approach)

n(WA)(w, a) =

and Newman’s normalization

n′(WA)(w, a) =

to temporal networks and using the temporal networks multiplication we finallyget the temporal cumulative co-authorship network

CNt = n(WAtc)T · n′(WAtc)

This is an undirected network with loops removed.We compute the weighted degree cores of the network CNt. It took 20

minutes on a standard PC. We get temporal weighted degree core numbers for93011 authors. Again, the obvious result is to display the most active authors.Since the values in TQs in cumulative co-authorship network are increasing wecan use as a measure of activity the accumulated fractional value in the lastyear. For g = 63 we get the list of authors in Table 3.

Figure 10 presents the heatmap for these authors.*** CommentsIf we would like to get more precise information about the TQs of selected

nodes we can present them as "histograms". In Figure 11 we compare the TQsfor the weighted degree cores for Borgatti and Pattison.

In (very) large networks most of the nodes are of small degree. If we are onlyinterested in nodes with large core numbers, say at least k, we can speed-up thecomputation by considering the property that all nodes in the k-core have degreeat least k. Therefore we can remove from the node degree data all intervals forwhich the node degree is less than k. We also remove all nodes with empty TQresulting from this transformation. We apply the temporal cores procedure onthe reduced network and apply the same reduction on the obtained temporalcore.

5 ConclusionTODO

16

Page 17: Cores in temporal networks A longitudinal approach

1976

1977

1978

1979

1980

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

BORGATTI_S

EVERETT_M

CHRISTAK_N

FOWLER_J

PATTISON_P

ROBINS_G

OHARA_K

BARABASI_A

LISBOA_E

COUTINHO_C

GRABOWSK_A

KOSINSKI_R

LATKIN_C

DAVEY−RO_M

MEYBODI_M

REZVANIA_A

BATAGELJ_V

DOREIAN_P

FERLIGOJ_A

MRVAR_A

LITWIN_H

STOECKEL_K

VALENTE_T

AGUILLO_I

ORTEGA_J

ARENTZE_T

TIMMERMA_H

STEINHAU_H

METZKE_C

TOBIN_K

KAZIENKO_P

BRODKA_P

NGUYEN_H

VASILAKO_A

NEWMAN_M

PATACCHI_E

KRAUSE_J

CROFT_D

JAMES_R

AGGARWAL_C

KANDHWAY_K

THOVEX_C

KURI_J

TRICHET_F

PRASANNA_V

CHELMIS_C

ABBASI_A

ALBERT_R

BERNARD_H

KILLWORT_P

STATTNER_E

COLLARD_M

LEYDESDO_L

VANDENBE_P

ALTMANN_J

BLACHNIO_A

CRANMER_S

DESMARAI_B

FAUST_K

WASSERMA_S

SKVORETZ_J

FARARO_T

GIANNAKI_G

5

10

15

20

Figure 10: SNA literature largest Ps cores.

Analyze the complexity of the algorithmImprove the complexity of the algorithmExtend the algorithm to generalized temporal coresFind user friendly presentations of resultsCompare with the streaming core algorithmsTemporal Quantities - a Python 3 library for temporal network analysis:

http://vladowiki.fmf.uni-lj.si/doku.php?id=tq

17

Page 18: Cores in temporal networks A longitudinal approach

1977

1978

1979

1980

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

0

5

10

15

20

25

BORGATTI_SPATTISON_P

Figure 11: SNA literature – comparing Borgatti and Pattison.

AcknowledgmentsWe thank the two anonymous reviewers whose comments and suggestions helpedimprove and clarify this manuscript.

All computations were performed using the program for large network anal-ysis and visualization Pajek [?] and the statistical programming system R.

This work was supported in part by the Slovenian Research Agency (researchprogram P1-0294 and research projects J1-9187 and J5-2557), and preparedwithin the framework of the HSE University Basic Research Program.

The second author was partially sponsored by Slovenian Research Agency(ARRS) - projects Z7-7614 (B).

Computational details (code) are available athttps://github.com/bavla/Cores/wiki/paper .

[19, 8, 5, 7, 18, 14, 16, 23]

References[1] A. Ahmed, V. Batagelj, X. Fu, S.-H. Hong, D. Merrick, and A. Mrvar. Vi-

sualisation and analysis of the internet movie database. In 6th InternationalAsia-Pacific Symposium on Visualization, pages 17–24. IEEE, 2007.

[2] H. Aksu, M. Canim, Y.-C. Chang, I. Korpeoglu, and Ö. Ulusoy. Multi-resolution social network community identification and maintenance on bigdata platform. In 2013 IEEE International Congress on Big Data. IEEE,2013.

[3] H. Aksu, M. Canim, Y.-C. Chang, I. Korpeoglu, and Ö. Ulusoy. Distributedk-core view materialization and maintenance for large dynamic graphs.IEEE Transactions on Knowledge and Data Engineering, 26(10):2439 –2452, 2014.

[4] S. Aridhi, A. Montresor, and Y. Velegrakis. Bladyg: A graph processingframework for large dynamic graphs. Big Data Research, 9:9–17, 2017.

18

Page 19: Cores in temporal networks A longitudinal approach

[5] V. Batagelj and S. Praprotnik. An algebraic approach to temporal networkanalysis based on temporal quantities. Social Network Analysis and Mining,6(28), 2016.

[6] V. Batagelj and M. Zaveršnik. Fast algorithms for determining (gener-alized) core groups in social networks. Advances in Data Analysis andClassification, 5:129–145, 2010.

[7] Vladimir Batagelj. On fractional approach to analysis of linked networks.Scientometrics, 123(2):621–633, May 2020.

[8] Vladimir Batagelj and Daria Maltseva. Temporal bibliographic networks.Journal of Informetrics, 14(1):101006, 2020.

[9] Vladimir Batagelj, Andrej Mrvar, and Matjaž Zaveršnik. Partitioning ap-proach to visualization of large graphs. In Jan Kratochvíyl, editor, GraphDrawing, volume 1731 of Lecture Notes in Computer Science, pages 90–97.Springer, Berlin / Heidelberg, 1999.

[10] Vladimir Batagelj and Matjaž Zaveršnik. Generalized cores. CoRR,cs.DS/0202039, 2002.

[11] Vladimir Batagelj and Matjaž Zaveršnik. An o(m) algorithm for coresdecomposition of networks. CoRR, cs.DS/0310049, 2003.

[12] G. Cantos-Mateos, M.A. Zulueta, B. Vargas-Quesada, and Z. Chinchilla-Rodriguez. Estudio evolutivo de la investigacion espanola con celulasmadre. visualizacion e identificacion de las principales l’ineas de investi-gacion. El Profesional de la Informacion, 23(3):259–271, 2014.

[13] M. Cerinšek and V. Batagelj. Generalized two-mode cores. Social Networks,42:80–87, 2015.

[14] Steven Corman, Timothy Kuhn, Robert Mcphee, and Kevin Dooley. Study-ing complex discursive systems: Centering resonance analysis of communi-cation. Human Communication Research - HUM COMMUN RES, 28:157–206, 04 2002.

[15] R. Franzosi. Mobilization and countermobilization processes: From the redyears (1919-20) to the black years (1921-22) in italy. a new methodologicalapproach to the study of narrative data. Theory and Society, 26(2-3):275–304, 1997.

[16] Edoardo Galimberti, Martino Ciaperoni, Alain Barrat, Francesco Bonchi,Ciro Cattuto, and Francesco Gullo. Span-core decomposition for temporalnetworks: Algorithms and applications. ACM Trans. Knowl. Discov. Data,15(1), December 2020.

[17] R.-H. Li and J. X. Yu. Efficient core maintenance in large dynamic graphs.IEEE Transactions on Knowledge and Data Engineering, 26:2453–2465,2014.

[18] Fragkiskos D. Malliaros, Christos Giatsidis, Apostolos N. Papadopoulos,and Michalis Vazirgiannis. The core decomposition of networks: theory,algorithms and applications. The VLDB Journal, 29(1):61–92, Jan 2020.

19

Page 20: Cores in temporal networks A longitudinal approach

[19] Daria Maltseva and Vladimir Batagelj. Social network analysis as a field ofinvasions: bibliographic approach to study sna development. Scientomet-rics, 121(2):1085–1128, Nov 2019.

[20] S. B. Seidman. Network structure and minimum degree. Social Networks,5:269–287, 1983.

[21] N. Wang, D. Yu, H. Jin, C. Qian, X. Xie, and Q.-S. Hua. Parallel algorithmfor core maintenance in dynamic graphs. In 2017 IEEE 37th InternationalConference on Distributed Computing Systems. IEEE, 2017.

[22] Stanley Wasserman and Katherine Faust. Social network analysis: Methodsand applications. Structural Analysis in the Social Sciences (8). CambridgeUniversity Press, 1994.

[23] Huanhuan Wu, James Cheng, Yi Lu, Yiping Ke, Yuzhen Huang, Da Yan,and Hejun Wu. Core decomposition in large temporal graphs. In 2015IEEE International Conference on Big Data (Big Data), pages 649–658,2015.

[24] Y. Zhang, J. Xu Yu, Y. Zhang, and L. Qin. A fast order-based approach forcore maintenance. In 2017 IEEE 33rd International Conference on DataEngineering. IEEE, 2017.

.

20