A Social IoT-driven Pedestrian Routing Approach during Epidemic Time
AA Social IoT-driven Pedestrian Routing Approachduring Epidemic Time
Abdullah Khanfor, Hamdi Friji, Hakim Ghazzai, and Yehia Massoud
School of Systems & Enterprises, Stevens Institute of Technology, Hoboken, NJ, USA
Abstract —The unprecedented worldwide spread of coronavirusdisease has significantly sped up the development of technology-based solutions to prevent, combat, monitor, or predict pandemicsand/or its evolution. The omnipresence of smart Internet-of-things (IoT) devices can play a predominant role in designingadvanced techniques helping in minimizing the risk of contam-ination. In this paper, we propose a practical framework thatuses the Social IoT (SIoT) concept to improve pedestrians safelynavigate through a real-wold map of a smart city. The objective isto mitigate the risks of exposure to the virus in high-dense areaswhere social distancing might not be well-practiced. The proposedrouting approach recommends pedestrians’ route in a real-timemanner while considering other devices’ mobility. First, the IoTdevices are clustered into communities according to two SIoTrelations that consider the devices’ locations and the friendshiplevels among their owners. Accordingly, the city map roads areassigned weights representing their safety levels. Afterward, anavigation algorithm, namely the Dijkstra algorithm, is applied torecommend the safest route to follow. Simulation results appliedon a real-world IoT data set have shown the ability of theproposed approach in achieving trade-offs between both safestand shortest paths according to the pedestrian preference.
Index Terms —Internet of Things (IoT), community detection,smart city, coronavirus, COVID-19, routing.
I. I
NTRODUCTION
In early 2020, the world was hit by an unprecedentedpandemic that severely affected most countries and createdglobal health care and economic pressures. To prevent itsspread, limiting the exposure to the virus is the main priorityof local authorities. Precautionary practices such as handcleaning, mask-wearing, social distancing, and close contactavoidance are highly recommended and even imposed in manycountries. Besides, technological solutions have been testedand implemented to help mitigate the spread of COVID-19 in the world. One of the most promising approachesis to exploit heterogeneous and omnipresent communicationsystems such as the Internet of Things (IoT) to enable e-monitoring techniques such as spread tracking, contact tracing,and crowded areas monitoring. IoT can provide cost-efficientand practical solutions to help practicing social distancing andhence, limit the spread of the infection [1], [2].
This paper is accepted for publication in 2020 IEEE Global Conference onAI & IoT IEEE (GCAIOT’20), Dubai, UAE, Dec. 2020.© 2020 IEEE. Personal use of this material is permitted. Permission fromIEEE must be obtained for all other uses, in any current or future media,including reprinting/republishing this material for advertising or promotionalpurposes, creating new collective works, for resale or redistribution to serversor lists, or reuse of any copyrighted component of this work in other works.
The current smartphones and wearable devices’ and theexisting infrastructure can boost the development of IoT-based solutions in a quick and large-scale manner to combatpandemics. Many examples of IoT-based solutions to contendthe pandemic effects have been proposed in literature [3][4] [5]. For instance, the smart disease surveillance systemshad demonstrated an efficient degree of control for the pan-demic’s spread within the city of Wuhan and other majorcities in China [6]. Despite the significant issues of privacy,South Korea’s exemplary accomplishments for containing theCOVID-19 until today is due, in part, to the commissioning acoherent information system that tracks visitors and confirmedpatients with an alerting system of potential infections [7]. Thesystem provides the community with essential information toassess the spread. Taiwan used various IoT technologies, suchas tracking the citizens and travelers through their mobilephone locations. Thus, if citizens are exposed to an areawith a high risk of getting infected, they will be altered.Also, if a traveler comes from a high-risk area and violatesself-quarantine procedure during the incubation phase, in thatcase, the residents in that area will be notified through a textmessage to alert them [8].A fundamental solution to reduce the transmission of in-fectious diseases in general and particularly the COVID-19is maintaining social distancing. IoT can perform a vitalfunction in helping with social distancing practices. Thus,the built-in capabilities of connected devices such as GPS,thermometer, and other sensors in the IoT system can helpin social distancing. For instance, in construction or industrialzones, wearables can be employed to maintain a safe distancebetween workers by generating alerts if social distancing isviolated. It also helps track the spread of the virus in casethat an infected person was present in the working area andhence, avoid the complete shutdown of the institution [7].The emergence of social IoT (SIoT) can be a valuable toolto leverage the traditional IoT systems and enable a betterunderstanding of the ubiquitous IoT network [9]. SIoT modelthe devices and users in the system with different socialrelations interconnecting the IoT devices. These relationshipscan be established between machine-to-machine, human-to-machine, and human-to-human connections [10] and transformthe IoT network into a socially connected network of devicesthat can be effectively analyzed using graph analytics toolssuch as community detection [11] and machine learning [12].By assessing the SIoT, providing new applications to battlethe virus spread can emerge and contribute to minimizing the a r X i v : . [ c s . S I] J a n andemic’s negative impacts.In this paper, we propose a smart navigation frameworkintending to determine for pedestrians safe routing to bypassareas where the risk of COVID-19 transmission is high. Inother words, the framework recommends a pedestrian walkingroute in which guarantees a social distancing and avoidingclose contacts. The proposed approach includes four steps:First, the framework identifies the IoT devices located in thearea of interest and then establishes social graphs interconnect-ing these devices using different social IoT relations. Then, theLouvain community detection algorithm is applied to the SIoTgraphs to determine different communities of IoT devices. Inour approach, we focus primarily on two social relations: adistance-based relation that identifies crowded/high-density ar-eas of IoT devices and a device friendship relation that allowslabeling streets where the user may possess a high chance ofmeeting a close friend. The third step is to compute differentscores representing each street’s safety level or segment of astreet in the area of interest according to the nearby detectedsocial communities. Finally, in the last step, the city map istransformed into a weighted undirected graph to which weapply the Dijkstra algorithm [13] in order to determine aroute characterized by a certain level of safety. A weightedbi-objective function balancing between the shortest and safestroutes is developed and implemented. The framework will thendeliver the trajectories to the user, e.g., via a mobile applicationfor the best route to follow to reach a destination. The proposedrouting approach takes into account the mobility of IoT devicesand may update the recommended route regularly by repeatingthe process, as mentioned earlier.II. P ROPOSED P EDESTRIAN R OUTING A PPROACH
In Fig. 1, we present a flowchart of the proposed navigationframework where the four steps are showcased. The objectiveis to recommend a safe route for a user connected to the serverthrough its IoT device, e.g., a smartphone. To this end, twoinputs are required to determine the route: the offline city mapand a data set containing the IoT device information such asthe device locations, and device owners. The framework willoutput a route from a starting point A to a destination B that isalready pre-defined by the pedestrian of interest. The first stepis a pre-processing step in which three graphs are generated: i)a graph representing the city road map, ii) a graph representingthe social relation reflecting the geographical relation of theconnected IoT devices called the Co-LOcation based relation(CLOR), and a graph identifying the friendship levels amongthe IoT devices, called the Social Friendship and Ownership-based Relation (SFOR). The second step aims to understandIoT devices’ social relations better and relax the problem’scomplexity by determining communities of socially connectedIoT devices. To this end, we propose to employ a communitydetection algorithm, namely the Louvain method [14], on theCLOR and SFOR graphs. This step will output communitieswith different risk levels of virus exposure. In the next step,step 3, we assign to every edge of the city map graph a weightmeasuring the traveled distance as well as the safety level of the corresponding street or segment of the street. Finally,the last step applies a graph routing algorithm that minimizesthe weights along the selected trajectory and determines thebest path to recommend to the pedestrian of interest. In thefollowing, we explain in detail each step of the proposednavigation framework. A. Step 1: Data Pre-processing and Graph Generation
In order to adequately manipulate the road network of thegeographical area of interest, we convert it into a graph wherethe vertices correspond to the intersections of the roads orconnections of two consecutive segments of roads. Indeed,we divide long roads/streets having lengths higher than athreshold length L th into multiple segments to each segmentof road is treated solely and might be later assigned differentweights. The number of segments for a road of length L road can be expressed as (cid:100) L road L th (cid:101) . Hence, the edges of the graphare the obtained segments of roads connecting two consecutiveconnections or the ones connected with an intersection.Besides, we generate two other graphs related to the differ-ent social relations and based on the available data set of theconnected IoT devices. In the following, we list two differentrelations that are used in this study to determine to measurethe social interconnections among the IoT devices: • Co-location/co-work based relation (CLOR):
The geograph-ical locations of the IoT devices can be used to define aspecific relation reflecting the fact that two devices are co-located in a given area at a certain instant of time. By settinga defined threshold for the distance between the devices, wecan specify whether these devices belong to a specific clusteror not and hence, establish relations between them based ontheir separating distances. This relation will allow identifyingcrowded areas where there is a risk that social distancing isnot practiced among these devices and hence, it is vital toavoid passing by these hot spots. • Social friendship and ownership relation (SFOR):
In thisrelation, we identify the devices that might be owned by thepedestrian of interest of his/her social friends. It is assumedthat there is more chance that the pedestrian will meet orbe in close contact with people using these devices, andhence, there is a risk of contamination with people that he/sheknows. To create SFOR relations, we first consider that twodevices owned by the same person are strongly connected.Regarding devices owned by different owners, we establishtheir SFOR relations using social media networks or otherfriendship indicators. For example, if two owners are friendsin the social network, then the SFOR relation between theirdevices can be modeled by an edge with a weight reflecting thestrength of their relations. SFOR relation can be extended tothe case of a friend of a friend with a reduced weight computedaccording to the number of friends needed to reach a specificdevice.The CLOR and SFOR topologies are undirected andweighted networks. The nodes in these graphs are heteroge-neous IoT devices. The edges between these devices repre-senting the SIoT relations stated previously. These graphs doig. 1: A proposed framework to recommend a safe and fast route to the user during a pandemic in SIoT.not include self-loop edges because the relations definitionsdo not require such a feature.
B. Step 2: Community Detection
This step focuses better on analyzing the social IoT re-lations, reducing the complexity of the problem, and servein information retrieval. A community detection algorithmconverts the complex social graphs into clusters of devicessharing strong relations. To this end, we apply the Louvainmethod [14]. The main advantage of using the Louvain isthe running time of O ( n log n ) , which is considerably fastercomparing to a similar methods [9], [15]. The outcomes ofthe community detection in our framework will be used in thenext Step 3 based on the relationships described in
Step 1 ,namely CLOR and SFOR.Applied to the CLOR graph, the Louvain method is ex-pected to extract co-located devices communities. Devices be-longing to the same community are considered to be positionednear one another and may create a highly dense zone thatis risky to cross through. In SFOR communities, the devicesare not necessarily co-located. On the contrary to CLOR, theymight be sparsely distributed in a geographical area. However,the owners of these devices may know each other and can meeteach other. Therefore, for safety reasons, it is recommended that a given pedestrian do not pass by devices belonging tothe same SFOR community of its device.
C. Step 3: City Map Edges’ Weights Computation
Since IoT devices are usually omnipresent, it is unlikelyto find routes free of devices, i.e., zero-risk zones. Therefore,in this step, we propose to compute weights and assign themto different edges of the road map graph given the statusesof their surrounding communities. Hence, the route selectionalgorithm will minimize the sum of weights along the route.In this step, we calculate the edges’ weights defined as afunction balanced by a coefficient α ( ∈ [0 , representingthe level of safety set by the pedestrian of interest. Settinga value of α → , the pedestrian intends to determine theshortest path to reach the destination with low consideration ofrisks. However, if α → , the pedestrian is looking to followthe safest trajectory independently of the expected traveleddistance. Values of α ∈ ]0 , achieves a trade-off betweenboth routing strategies. The edge’s weight of the road network,denoted by ω e can be expressed as follows: ω e = (1 − α ) ω diste + α ω sfte , (1)Where ω diste is the weight reflecting the expected traveleddistance when crossing edge e , while ω sfte is the weightreflecting the safety level of the edge. The weight ω diste is a a) CLOR communities (The colors of each CLOR commu-nity indicates the density level of the community). (b) SFOR communities (IoT devices belonging to the samecommunity are labeled by the same marker). Fig. 2: Communities were detected based on CLOR and SFOR relations using the Louvain Method.normalized value of its length. On the other hand, the value of ω sfte is obtained by combining the impact of the surroundingCLOR communities and devices belonging to the same SFORcommunity of the pedestrian’s device as follows: ω sfte = ω CLORe + ω SF ORe . (2)The CLOR weights ω CLORe are calculated using the followingexpression: ω CLORe = (cid:88) c ∈ C CLORe γ c |C CLORe | , (3)Where C CLORe is the set of CLOR communities that intersectwith the edge e . The cardinality of this set is denoted by |C CLORe | . The CLOR communities are modeled as polygonsthat circumscribe all devices belonging to them with an outeroffset ρ . We denote these polygons by P c , ∀ c ∈ C CLOR where C CLOR denotes the set of all CLOR communities obtainedusing the Louvain method. Hence, the set C CLORe can bedefined as: C CLORe = { c ∈ C CLOR | P c ∩ { e } (cid:54) = ∅} where ∅ is the empty set. Note that the offset parameter ρ is addedto all polygons associated with the CLOR communities toensure a safe social distance separating the navigating userand the devices at the edges. Finally, γ c denotes the densityof community c and is calculated as follows: γ c = N c A c , (4)Where N c is the number of devices in community c and A c is the surface of the area of P c .Similarly the SFOR weights can be computed as follows: ω SF ORe = (cid:88) u ∈C u ∗ ,e Ω SF ORu,u ∗ |C u ∗ ,e | , (5)Where C u ∗ ,e is the set of devices that are in SFOR relationwith the device u ∗ of the pedestrian of interest and their d ( u, e ) ≤ d th . It corresponds to the SFOR community towhich the device u ∗ belongs having a distance d th or less from the edge e . The coefficient Ω SF ORu,u ∗ is obtained fromthe SFOR graph, and it measures the SFOR relation betweendevice u and u ∗ . The parameter d th can represent a distancefrom which the users owning the IoT devices cannot see eachother and hence, do not meet and avoid close contact.Notice that the safety weight of an edge e significantly in-creases if it is surrounded by high-density CLOR communitiesand/or many devices belonging to the same SFOR communityof the user of interest. Therefore, in the next step, we aim toselect the edges, i.e., the trajectory that minimizes the sum of w sfte for a user looking for a safe walk. D. Step 4: Trajectory Recommendation
After computing the weights of the city map graph in
Step 3 , we apply the Dijkstra’s shortest path algorithm todetermine the trajectory with minimum cumulative weightsof the selected trajectory between the points A and B . Wepropose to employ Dijkstra’s algorithm due to its reasonablerunning. However, similar algorithms can be applied in ourcontext.Our framework is capable of dynamically recommendingnew paths to the user based on his/her current location whileconsidering the mobility of other IoT devices. Hence, it needsto update the CLOR and SFOR communities after a specificperiod. Consequently, the selected path is updated from a timeslot to another. In other words, the steps 2 to 4 will be repeatedfor each time slot until the user reaches his/her destination.III. R ESULTS & D
ISCUSSIONS
In our simulations, we select a × km area in Santander,Spain. We extract the map using OpenStreetMap project andconvert it to a road graph using the OSMnx method [16].Moreover, we project the devices in the selected area from a ig. 3: Two examples showing the different paths recommended to the user for different values of α .real-world IoT data set provided in [17]. The data set includes16216 devices covering the whole city. The devices vary fromsimple sensors such as street lights, environment sensors, andhighly computational devices such as smartphones and per-sonal computers. The devices are owned by private and publicentities. The local authorities usually own public devices. Forthe private-owned devices, there are static and mobile devices.In our simulations, we select mobile devices owned by privateentities that are most likely owned by human beings such assmartphones, smartwatches, tablets, personal computers, etc.The remaining devices from the previous selection processresult in 1312 personal IoT devices.The selected devices will have SFOR and CLOR SIoTrelations. For the CLOR relations, we create a mesh networkand drop the edges connecting two devices separated by adistance higher than 1 km. The community detection algorithmapplied to the CLOR will return a set of relatively high-densitycommunities to determine high-risk infection areas with thelimited practice of social distancing. For the SFOR relation,we employ the social network of the owners of the IoTdevices. Since we lack access to the owners’ social network,we use Watts–Strogatz generator [18] that portrays a socialnetwork between the owners. The relations of devices withthe same owner are assigned an edge of 1, while the directfriends-owned devices will have an edge of 0.5. Other devicesare given weights computed while considering the minimumnumber of hops needed for one of the vertices (owners) toreach the other vertices. We restricted the relations to threefriends of friends since they are socially far away from eachother.Fig. 2 shows the communities obtained from the SIoTgraphs by applying the Louvain method. We obtain 56 CLOR communities represented by colored polygons and havingdifferent density levels, as illustrated in Fig. 2a. The CLORcommunities are classified, based on their densities, into fiveclasses. There is one very high-density community locatedalmost at the center of the map and other blue CLOR com-munities with high-density that the user needs to avoid forsafety. The medium and low-risk areas might be avoided, butthey can be recommended. For the SFOR, it results in 10communities with diverse types of devices located all over themap. Each community is denoted with different shapes andcolors in Fig. 2b. The user of interest will belong to one ofthese communities and needs to avoid close contact with them.Fig. 3 illustrates two examples of the recommended routesfor the user given different starting points and destinationsfor three values of α after applying the Dijkstra’s algorithmusing the computed weights. If α = 0 , then the frameworkwill recommend the shortest path (the red route), otherwise if α = 1 , the safest path with minimum exposure to the virus isrecommended (the green). However, for α = 0 . , a trade-offbetween both metrics is provided (the blue route). In Fig. 3, itis clear that along the green route, the user is avoiding mostof the high dense areas by surrounding them. It just crossessome of the low-density areas. It also avoids getting closer toother SFOR-related devices unless it is forced to do it. Thisleads to a long route of 2.77 km. With the red route, the user isunaware of the risk of contamination and crosses all the high-density areas. The corresponding traveled distance is equal to2.1 km. Finally, for the blue route, the algorithm avoids thered zone and tolerates passing by some blue areas. This resultsin a traveled distance of 2.4 km.Our framework can be adapted to a dynamic scenario byexamining real-time mobility. In Fig. 4, we plot the routes forig. 4: The routing updates on three timesteps are based on the changes in device locations and communities in the graph.Fig. 5: A trade-off between the safety factor and destinationamount for the proposed framework.three consecutive time slots. Each time slot, the IoT deviceschange their locations and, consequently, influence the CLORrelations and their communities and the position of devicesin SFOR. In the dynamic scenario, the shortest path willremain intact. The algorithm is only aware of the distanceto be crossed, while in the proposed framework that considersthe safety weights, the trajectory is regularly updated giventhe location of the devices at each time slot. Accordingly,the user starting at the left bottom corner of the map willnotice that its trajectory is partially updated at a time slot ( t )since several SFOR devices left the area, and the user cancross in the middle of the map to reach its destination. Inthe next time slot ( t ), the navigation algorithm is executedagain. Notice that the user is forced to go around it to reach itsdestination. As long as he/she is moving, the user is gettingcloser to the destination, especially if a correct value of α is chosen. Choosing α close to 1 may lead to confusingresults. Therefore, to balance between safety and travel time,an optimized choice of α should be made. In Fig. 5, we plot the final travel distance and a safetyscore measuring the cumulative safety weights along thetrajectory versus different values of α for two choices of ρ (thesocial distancing outer offset). A higher value of ρ indicatesan increasing preventive navigation strategy aiming to findtrajectories a little bit far away from CLOR communities. Thefigure shows that by increasing α , the travel distance increases,moving from 1.7 km to around 2.5 km while the cumulativesafety score is almost linearly decreasing. A compromisebetween safety and speed can be achieved for α around 0.4.By increasing the outer offset ρ , a more strict social distancingis applied, and hence, the traveled distance increases even forthe same value of α . For instance, for α = 0 . , the distancechanges from 1.95 km to 2.21 km, with a slight improvementof the safety score. IV. C ONCLUSION
Ubiquitous IoT can provide solutions to combat pandemicssuch as COVID-19. We provide a practical framework thatcould assist in such circumstances. The framework recom-mends a trajectory for a pedestrian user to reach his/herdestination while avoiding areas with high risk of exposure tothe virus. It employs technical and social advantages of smartIoT devices to determine risky areas and help people betterpractice social distancing. The framework can be update therecommended route in real-time based on the user’s needs andmobility of other devices. As future work, the framework willbe extended to consider the case of multi-user routing andcan be adapted to more complicated areas such as indoor andindustrial workplaces. R
EFERENCES[1] R. P. Singh, M. Javaid, A. Haleem, and R. Suman, “Internet of things(iot) applications to fight against covid-19 pandemic,”
Diabetes &Metabolic Syndrome: Clinical Research & Reviews , vol. 14, pp. 521– 524, July-Aug. 2020.[2] M. S. Rahman, N. C. Peeri, N. Shrestha, R. Zaki, U. Haque, and S. H. A.Hamid, “Defending against the novel coronavirus (covid-19) outbreak:How can the internet of things (iot) help to save the world?,”
HealthPolicy and Technology , vol. 9, pp. 136 – 138, June 2020.3] V. Chamola, V. Hassija, V. Gupta, and M. Guizani, “A comprehensivereview of the covid-19 pandemic and the role of iot, drones, ai,blockchain, and 5g in managing its impact,”
IEEE Access , vol. 8,pp. 90225–90265, May 2020.[4] A. Pimpinella, A. E. Redondi, and M. Cesana, “Walk this way! aniot-based urban routing system for smart cities,”
Computer Networks ,vol. 162, p. 106857, Oct. 2019.[5] M. Hatzopoulou, S. Weichenthal, G. Barreau, M. Goldberg, W. Farrell,D. Crouse, and N. Ross, “A web-based route planning tool to reducecyclists’ exposures to traffic pollution: A case study in montreal,canada,”
Environmental Research , vol. 123, pp. 58–61, May 2013.[6] N. C. Peeri, N. Shrestha, M. S. Rahman, R. Zaki, Z. Tan, S. Bibi,M. Baghbanzadeh, N. Aghamohammadi, W. Zhang, and U. Haque,“The SARS, MERS and novel coronavirus (COVID-19) epidemics, thenewest and biggest global health threats: what lessons have we learned?,”
International Journal of Epidemiology , vol. 49, pp. 717–726, Feb, 2020.[7] D. S. W. Ting, L. Carin, V. Dzau, and T. Y. Wong, “Digital technologyand covid-19,”
Nature medicine , vol. 26, pp. 459–461, Mar. 2020.[8] C. J. Wang, C. Y. Ng, and R. H. Brook, “Response to covid-19 intaiwan: big data analytics, new technology, and proactive testing,”
Jama ,vol. 323, pp. 1341–1342, Apr. 2020.[9] A. Khanfor et al., “Application of community detection algorithms onsocial internet-of-things networks,” in
IEEE International Conference onMicroelectronics (ICM’19) , pp. 94–97, Cairo, Egypt, Dec. 2019.[10] L. Atzori, A. Iera, G. Morabito, and M. Nitti, “The social internet ofthings (SIoT)–when social networks meet the internet of things: Concept,architecture and network characterization,”
Computer networks , vol. 56, pp. 3594–3608, Nov. 2012.[11] J. Shuja, K. Bilal, E. Alanazi, W. Alasmary, and A. Alashaikh, “Ap-plying machine learning techniques for caching in edge networks: Acomprehensive survey,” arXiv preprint arXiv:2006.16864 , 2020.[12] A. Khanfor, R. Hamadi, H. Ghazzai, Y. Yang, M. R. Haider, andY. Massoud, “Computational resource allocation for edge computingin social internet-of-things,” in , pp. 233–236, Aug.2020.[13] E. W. Dijkstra, “A note on two problems in connexion with graphs,”
Numerische Mathematik , vol. 1, pp. 269–271, Dec 1959.[14] V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and E. Lefebvre, “Fastunfolding of communities in large networks,”
Journal of statisticalmechanics: theory and experiment , vol. 2008, p. P10008, Oct. 2008.[15] A. Khanfor, H. Ghazzai, Y. Yang, M. R. Haider, and Y. Massoud,“Automated service discovery for social internet-of-things systems,”
IEEE International Symposium on Circuits and Systems (ISCAS’20),Seville, Spain. , Oct. 2020.[16] G. Boeing, “Osmnx: New methods for acquiring, constructing, analyz-ing, and visualizing complex street networks,”
Computers, Environmentand Urban Systems , vol. 65, pp. 126–139, May 2017.[17] C. Marche et al., “A dataset for performance analysis of the social in-ternet of things,” in
IEEE Annual International Symposium on Personal,Indoor and Mobile Radio Communications (PIMRC’18) , Bologna, Italy,Sept. 2018.[18] D. J. Watts and S. H. Strogatz, “Collective dynamics of ‘small-world’networks,” naturenature