Homophily-Based Network Formation Models

Size: px
Start display at page:

Download "Homophily-Based Network Formation Models"

Transcription

1 Homophily-Based Network Formation Models Senior Honors Project Final Report Scott Linderman School of Electrical and Computer Engineering Cornell University Frank H T Rhodes Hall Ithaca, NY swl28@cornell.edu ABSTRACT Many network formation models have been proposed to explain the characteristics of real world networks which cannot be captured by random graphs. Though random graphs do have a small diameter, they do not exhibit the high clustering and heavy-tailed degree distributions that have been observed in several real world networks. The Watts-Strogatz Small World model produces graphs with high clustering and the Barabasi-Albert Preferential Attachment model can create the power-law degree distribution of some real networks, but neither exhibits both properties alone. I believe one reason why these models do not acheive both realistic properties is that they do not incorporate the notion of homophily. The goal of this project was to explore network formation and evolution when the similarity of nodes is taken into account when forming edges. I examine many of these models and attempt to build similarity proportional neighbor selection into current models. I show one model which successfully exhibits both realistic properties. Keywords Networks, Network Formation Models 1. INTRODUCTION Homophily, the tendency for similar individuals to associate, is a known characteristic of real world networks. This tendency manifests itself in networks through clusters and communities. Its influence on the structure of networks makes it a key factor in how information or contagions spread on the network. Despite the importance of homophily, it is either absent or implicitly assumed to be uniform random in most prominent network formation models. The goal of my research project has been to analyze and experiment with network formation models with heterogeneous nodes where links are formed with some regard to the similarity of the two connecting nodes. This project has undergone many mutations, but it has always revolved around the idea of creating networks that exhibit the measured properties of real world networks while requiring similar nodes to be short distances from each other. The motivating example is a network of music listeners who desire to find new music that they enjoy, and in order to do so they associate themselves with others who have similar tastes. With this example in mind I experimented with many different models beginning with very detailed models of recommendation networks and eventually culminating with an abstract variant of prominent network formation models. In the final stage of this project I attempted to create a model which incorporates both the power law degree distribution and the high clustering that are often observed in real world networks but are not achieved by the standard models. The results of this work show that the concept of homophily is an important aspect of real networks, and that by modifying current network formation models to incoporate this idea we can create networks with more realistic properties. 2. PREVIOUS WORKS There has been a great deal of recent research on network formation algorithms, especially on models which have an aspect of randomness involved in link formation. Erdos and Renyi s seminal 1960 study [3] introduced the concept of random graphs in which each edge is formed independently with given probability. Though not necessarily realistic, this model offers interesting insight into the connectedness and degree distribution of networks. It was found that even randomly generated graphs are still connected and have a small diameter when the probability of forming an edge exceeds a certain threshold. Though random graphs have some nice attributes, they do not exhibit some of the fundamental characteristics of real networks, one of the most important being a large amount of clustering. Watts and Strogatz introduced the Small World Model [6] in 1998 to explain the formation of networks with small diameter and high clustering. Their model begins with a graph with complete local connections and then rewires a small fraction of the local connections to randomly chosen nodes. This retains high clustering,independent of the number of nodes, while also significantly reducing the average distance between nodes. Another model was introduced by Barabasi and Albert (BA) to explain the observed power law degree distributions in

2 some real world networks [2]. Their process of preferential attachment forms networks over time, a process they call growth, where each new node creates the same amount of edges, but chooses the destination vertex with probability proportional to the vertex s current degree. This favors older nodes and promotes the rich get richer concept. It was emperically determined, however, that the clustering coefficient in this model still decays in power-law fashion as the number of nodes increases. Holme and Kim proposed a model in 2002 [4] which attempted to exhibit both high clustering and a power law degree distribution. Their model explicitly allows for triadic closure to occur in an attempt to raise the level of clustering in the network. Holme and Kim achieve this by following each preferential attachment step of the BA algorithm with a Triadic Formation step. In this combined process, each new node enters the network, selects its neighbors first based on current degree, and finally randomly closes the triangles with its neighbors with a given probability. In this way, Holme and Kim are able to achieve networks with power law degree distribution with tunable clustering, based on the probability of closing triads. Though their model creates networks with both desireable characteristics, their method seems somewhat forced and unrealistic. Some attempts have been made to explain the real-world mechanisms which lead to the observed preferential attachment behavior. One of the more provactive explanations is the copying mechanism from Kleinberg et. al [5]. In this model each new node chooses a random partner to copy called the prototype. The new node then chooses each of its remaining edges by copying one of the prototype s edges with probability 1 p and choosing a random vertex with probability p. This has been shown to produce a power-law distribution of degree in a method which is equivalent to linear preferential attachment. There have been many modifications of these algorithms as well, and one of the most pertinent to this study was the Bianconi-Barabasi fitness model [1]. They modified the preferential attachment model by multiplying each node s probability of being chosen by a fitness parameter. This allows even young nodes to acquire edges quickly if they have a high relative fitness. Though this model differentiates nodes with a notion of valuability, it assumes that the value of a node is the same to all others. This is not the case in my models; rather than nodes having an inherent fitness, value is derived from similarity. 3. ATTEMPTED MODELS This project has evolved over time as various attempted models provided new insights and led to subsequent revisions. The evolution followed three main stages: I first attempted to simulate a complex network of nodes exchanging recommendations and reevaluating their neighbors based on the accuracy of their recommendations; then I tried to simplify the model by using difference equations to update the weights of an adjacency matrix; finally I attempted to incorporate homophily into three existing network models. These stages follow a trend towards simpler model with fewer parameters. I found that the more complex models were too difficult to analyze, and even though the simpler models may sacrifice some realism, their predictability makes them useful tools. 3.1 Recommendation Networks My first experiments involved a network of heterogeneous agents in a world with a finite number of valuable items. Each player had a ranking of these items, modelled as a sequence of the items, and initially they only were aware of an ordered subsequence of that ranking. In each time step the agents become aware of a previously unknown item, chosen based on the recommendations of their neighbors. With each recommendation the agents recalculate their pairwise valuations and then choose which links to form or sever. As players are exposed to more items their valuations evolve as do their choices of neighbors, thereby creating a system of simultaneous network formation and information diffusion. The model I used was fairly complex and involved very specific, hand-crafted functions for estimating the value of another player. To avoid the tedious mathematical details, the valuation function v ij can be summarized as a first-order linear predictor of the accuracy of j s recommendations to i used to scale the expected value of j s future rankings. This expected value is measured by finding a least-squares mapping of ranks of items in j s ranking to ranks for i, then finding the average predicted value of items which j could recommend. In each time step players receive recommendations from all of their neighbors and choose one recommendated item to be exposed to. This choice is made by considering the value of the player who made the recommendation, as well as the number of recommendations received for the item. For a more thorough treatent of this approach refer to the midterm report. The obvious complexity of this model makes it difficult to analyze and understand. Though I did see increasing tendency to link with similar neighbors and a slight increase in the value of the received recommendations, the results were very sensitive to the choice of parameters and not quite as expected. Rather than continuing with this model, I decided to switch to a simpler method with fewer variables. 3.2 Weight Matrix Update Rules In order to alleviate some of the complexity of the recommendation network I decided to model the network as a weighted directed graph with weight matrix W. These weights w ij were updated by two functions. First, I used a linear predictor to estimate the expected value of j s next recommendation based on previous recommendations values. Second, I added the i, jth element of the squared weight matrix W 2. The second component incorporates weight diffusion through the network based on the mantra a friend of my friend is also my friend. Recommendations were accepted from neighbors whose weight exceeded a given threshold, and items were chosen from among these recommendations in the same manner as before. Using the weight matrix eliminated the least squares predictions to value players and instead relied solely on past recommendation values and the neighbors value of the player. Though this model offered some nice simplifications to the original model, it turned out to have troublesome cases where the total weight of the matrix would increase without bound.

3 Normalizing the weights for each player prevents this symptom, but I believe the fundamental problem is that the linear predictor is not bounded and does not necessarily converge. Having found the recommendation framework and its requisite design copmlexities too unpredictable cope with, I switched course and defined similarity as Euclidean distance in feature space. This definition is far easier to analyze because it is deterministic and does not require sensitive definitions of how recommendations are made and valuation functions updated. The simplest realization of this similarity is a using a onedimensional feature. We give each node a feature value f i which is a real number on the range [0, 1]. Then, the similarity of two nodes i and j is defined as s ij = 1 f i f j (1) (a) This feature can be chosen according to any distribution, and obviously different distributions will yield different network structures. Unless otherwise stated, the following models assume the feature is drawn according to the uniform distribution on [0, 1]. The first model I used with this definition of similarity continued along the lines of the weight matrix update rules. Initially all nodes began in a small world graph with all edges equally weighted. Then at each time step each node chose one neighbor to associate with, resulting in one or more connected components of nodes. These neighbors are chosen with probability proportional to their similarity. Once these groups are chosen, each player sets the weight of those in his group equal to their similarity. The value of all players not in the group is depreciated by a constant factor β. Thus { sij j C w ij = i otherwise βw ij Figure 1: Figure (a) shows an evolution of groups in a simulation with 50 nodes. These snapshots are taken every 10 steps and the various colors show various groups. As you can see, over time the groups become more sparse, and there are more groups. This is because nodes become more likely to associate with similar neighbors. Figure shows the average similarity as defined above over time. This result is the average of 10 runs and you can see that similarity does indeed increase with time as nodes begin to exhibit homophily. where C i is the set of nodes in i s group. This update rule was designed to emphasize connecting to the most similar others and I expected that by choosing the node to associate with based on their similarity eventually groups of similar nodes would begin to appear. I found that the average similarity between nodes defined as V = 1 N N W [n] S[n] n=0 is initially around 0.7 starting in a small world, and that this model it usually increased in an oscillatory fashion towards 0.8 or even higher. When groups are chosen proportional to similarity I found that the highest weighted node for a given player oscillates around approximately 0.25 probability of being chosen, and that is not necessarily the most similar player in the entire world. This seems to indicate that the update rule is not discriminating enough against worse nodes. Perhaps over time the willingness to associate

4 with less similar nodes should decrease, just as in the simulated annealing algorithm. When every node chooses to associate with the highest weighted node, however, the average similarity approaches 1 very quickly. This, however, is a product of nodes finding their most similar neighbors and continually associating with them resulting in many groups of 2. Though this model has some promise, I doubt it will be able to create the power law degree distributions that we would expect from real social networks. Rather than continue experimenting with this model I attempted to incorporate the similarity function described above into some of the prominent network formation models since those already exhibit some desirable characteristics. 3.3 Homophily in Current Models I learned from my earlier experiments that a practical network formation model should not only exhibit the properties of the system it is trying to replicate, but should also achieve this goal using as few parameters as possible without sacrificing plausibility. With this in mind I attempted to incorporate my similarity function into three well-known network formation models: Watts and Strogatz s small world model, the Barabasi-Albert preferential attachment model, and Kleinberg s version of the prefential attachment model with the copying mechanism. There is some implicit notion of homophily that could be assumed in some current models. For instance, the ring lattice used to initially connect nodes in the Watts-Strogatz model could be interpreted as a measure of similarity where like nodes are positioned near each other on the ring. The Preferential Attachment model, however, does not have an obvious analogy for similarity based connections. I would argue that it is the notion of homophily that causes dense communities to form, and that the Watts-Strogatz model s inherent inclusion of this idea is what allows it to create dense networks with high clustering. Though Preferential Attachment creates networks with slightly higher clustering than random graphs of the same size, it has been empirically shown that the clustering coefficient still decreases with the number of nodes according to a power law. By modifying Kleinberg s variant of the Preferential Attachment model to explicitly factor similarity into link formation I create a model which exhibits high clustering as well as a heavytailed degree distribution Valuation Function In the following sections I define the total value of node j to node i as v ij = α s ij d k s + (1 α) j ik k d k where d j is the degree of vertex j and α is a constant on the range of [0,1]. Thus the value of a node is based on both its relative similarity as well as its relative degree and the tradeoff is tunable with parameter α Cost Threshold based on SW (2) I interpreted the ring lattice used to connect nodes in the Watts-Strogatz model as a result of connecting nodes to neighbors less than a certain distance away in similarity space. As a logical extension to this method, I imagined positioning nodes in a metric space which incorporates both similarity and connectivity, then followed the same approach and connected nodes to nearby neighbors. I used the valuation function above to measure similarity and connected to nodes whose value exceeded a given cost threshold. I began with all the nodes unconnected. Then at each time step every node connected to the node with the highest value, as long as the value exceeded some cost threshold. I found that the cost threshold had to be kept quite low in order for any connections to form initially, and then once a node gained enough edges to exceed the cost threshold with connectivity alone, it rapidly accumulated edges from every other node. This is not a surprise since in each time step all nodes connect to the maximum valued other, and an increase in one node s degree causes all others connectivity value to decrease. Adding an element of randomness, perhaps choosing nodes to connect to with probability proportional to their value, would most likely produce a more even degree distribution, but this does not fit as nicely with the cost threshold idea Preferential Attachment In order to work around the cost threshold sensitivity I found in the small world modifications I decided to try modifying the preferential attachment model instead. In my version of the model new nodes are created every time step. Upon creation they choose a parameter m nodes to connect to. I chose the new neighbors using fitness proportional selection, so the probability of a new node i choosing j as a neighbor is P i(j) = vij k v ik I found that this model yielded the expected power law degree distribution, but it actually had a negative affect on clustering. This was because the similarity aspect of the valuation function did not place enough emphasis on the most similar nodes. As nodes accumulated more neighbors, the probability of actually choosing a node and its neighbor quickly approached 0, even when similarity was taken into consideration. Furthermore, older nodes (those which were created first) quickly accumulated degree greater than m, so even if a new node connected to an old node, it could not possibly connect to all of his neighbors as well. This is one inherent feature of preferential attachment models which prevents perfect clustering. Nevertheless, I found that Kleinberg s variant of preferential attachment, which I call the copying mechanism, allows for much higher clustering than the normal version Holme and Kim I incorporated homophily into Holme and Kim s model of preferential attachment with a triadic closure step by choosing the original neighbors with similarity proportional selection, just as was done in the version of the Barabasi-Albert

5 Figure 3: This figure shows the clustering coefficient using the modified version of the Holme and Kim preferential attachment model for various values of p, and the effect of varying α for each p. As the probability of closing triads,p, is increased the clustering increases as expected. It appears that increased reliance on similarity actually decreases the observed clustering in these experiments, though I do not have a good explanation for why this occurs. Proof of the power law degree distribution is omitted but can be found in Holme and Kim s paper. preferential attachment model. I obtain the same results as Holme and Kim - the clustering does increase with larger values of p - but the inclusion of the similarity measure had the opposite effect of what I intended. Holme and Kim show that for larger values of N the clustering approaches an equilibrium value proportional to p. Though this method successfully creates graphs with high clustering, I felt that the method of achieving these results was rather contrived. The Kleinberg model of web formation, which I have referred to as the copying mechanism, is a more realistic model which actually exhibited higher clustering as well Copying Mechanism In order to retain the realistic characteristics of preferential attachment, my model begins with Kleinberg s copying mechanism. A new node is introduced at every time step and again chooses a prototype node to copy. Rather than choosing this prototype randomly, however, I choose based on a convex combination of the vertex s degree and the similarity of the new node. Prototype vertices are chosen in proportion to their value, the same way that new neighbors were chosen in the preferential attachment-based model. As in the preferential attachment model, the algorithm starts with m 0 unconnected vertices. At each time step a new node is added and chooses a prototype to copy based on the scheme described above. Each incoming node connects to m m 0 others, including the prototype. Each of the other vertices is chosen by either copying one of the prototype s neighbors with probability 1 p or choosing a random neighbor with probability p. Varying the parameter α of the value function from 0 to 1 varies the weight on connectivity versus similarity. When α = 0 protoypes are chosen based solely on connectivity and the model is very similar to the preferential attachment model. If α = 1 incoming nodes choose prototypes irregardless of connectivity and the resulting model lacks the powerlaw degree distribution. Intermediate values of α exhibit both heavy-tailed degree distributions and high clustering. Experimental results show that these intermediate values of α produce clustering which is much higher than in the original preferential attachment models with the same number of nodes. This clustering still decays with increasing number of nodes, but it does so much more slowly than in the first model, the Preferential Attachment model. Choosing prototypes with emphasis on connectivity results in the high degree nodes being chosen, and if the prototype s degree is larger than the number of edges the incoming node creates then there will be many possible triads of nodes which cannot all connect. When emphasis is placed on connecting to similar nodes the probability of connecting to high degree nodes is lessened and more possible triads are closed, thus the clustering coefficient is greater for larger values of α (i.e. when the emphasis on homophily is stronger). If we examine the copying mechanism alone, when prototypes are chosen uniform randomly, we see that the original model also produces relatively high clustering. The clustering for the original model can be seen in plot of Figure 4 for various values of p. For p = 1, as in the simulations of the Copying Mechanism with similarity model above, we see that clustering is still much larger than in the original Preferential Attachment model and approximately the same as in the similarity enhanced version of the Copying Mechanism for α = 1. These results are very dependendent upon p, the probability of copying the prototype s neighbors. When this probability decreases and more random neighbors are chosen, the clustering decreases quickly. Copying neighbors is essential to creating high clustering because it connects triads of nodes. Randomly choosing neighbors is unlikely to produce the same effect. 4. CONCLUSIONS Although the present results seem somewhat varied and inconclusive, I feel there is still a role for homophily in network formations models. In this project I have explored many different ways of generating networks with nodal similarity. I ve examined hand-crafted models as well as logical extensions of established network formation models with homophily considered during the edge selection process. From the mixed results, I feel a few positive highlights have emerged. My experiments yielded new insights into how similarity can affect network formation. These results show that strong group formation can emerge, it just requires the right param-

6 (a) (a) Figure 2: Figure (a) shows the resulting degree distribution for various values of paramete α. As you can see, when α approaches 0 (complete emphasis on connectivity) the model is equivalent to the original preferential attachment model and the power law is maintained. As α is increased and similarity is more heavily valued, the degree distribution tends towards a more exponential form with fewer nodes accumulating large degree. Figure shows that the clustering did not increase for larger values of α as I had hoped. Connecting with similarity proportional selection did not cause higher clustering in this model. Figure 4: Figure (a) shows the affect of varying α on the clustering coefficient vs N for m = 20 and p = 1. The clustering coefficient still seems to follow a power-law type of relationship initially, but due to computational constraints issues I was unable to examine the behavior of the clustering coefficient for larger values of N. Small world models with the same m exhibit clustering coefficient equal to approximately 0.5 and independent of N. For higher values of α we see clustering coefficients which are comparable for the range of N examined. Figure shows the affect of varying α on the degree distribution for m = 20 and p = 0. The log-log plot shows the linear relationship characteristic of the power law distribution. Intermediate values of α retain the power law form but eventually sole reliance on similarity results in a distribution which decays approximately exponentially. The number of nodes with degree less than m needs further investigation.

7 eters to simultaneously see the traditional network features like power law degree distributions and high clustering. (a) Figure 5: Figure (a) shows the affect of varying p for fixed α in the Copying Mechanism model with similarity proportional prototype selection. As expected, the clustering shows a marked decrease over all numbers of nodes as the probability p of copying the prototype is decreased. The largest difference is between p = 1.0 and p = 0.75, which seems to indicate that even a slight probability of choosing random neighbors can severely decrease the clustering. Figure shows the results of the same experiment run on the original Copying Mechanism without similarity. Notice that the clustering decreases at a faster rate in the original model which seems to indicate that the similarity component can indeed increase clustering. Many aspects of my models could be studied further. In particular, the role of similarity is directly affected by the distribution of the characteristic feature value in the population of nodes. Experimenting with the homophily models under different attribute distributions should give rise to different results. For example, if the nodal attribute is drawn from a bimodal distribution, under the right parameters, I would expect to see two distinct groups form when the weight on similarity is high. Other interesting experiments would be to shift the population s average characteristic value over time to simulate a change in general opinion of the population. The range of possible extensions with this model is unlimited. Finally, the effect of homophily in other network formation models needs to be studied. I selected the models to emulate based on their performance in achieving other desireable network properties, but adding nodal similarity to other models should also produce intersting and valuable results. Although there are many more avenues through which homophily could be introduced to network formation models and much work which could be done, these results show promise and hopefully more study will reveal further insights into role of homophily in network formation. 5. REFERENCES [1] R. Albert and A. Barabási. Statistical mechanics of complex networks. [2] A. Barabasi and R. Albert. Emergence of scaling in random networks. Science, 286: , [3] P. Erdos and A. Renyi. On the evolution of random graphs. Publication of the Mathematical Institute of the Hungarian Academy of Sciences, 5:17 61, [4] P. Holme and B. J. Kim. Growing scale-free networks with tunable clustering. Phys. Rev. E, 65(2):026107, [5] J. M. Kleinberg, R. Kumar, P. Raghavan, S. Rajagopalan, and A. S. Tomkins. The Web as a graph: Measurements, models and methods. Lecture Notes in Computer Science, 1627:1 17, [6] D. Watts and S. Strogatz. Collective dynamics of small-world networks. Nature, 393(6684): , June 1998.

Models of Network Formation. Networked Life NETS 112 Fall 2017 Prof. Michael Kearns

Models of Network Formation. Networked Life NETS 112 Fall 2017 Prof. Michael Kearns Models of Network Formation Networked Life NETS 112 Fall 2017 Prof. Michael Kearns Roadmap Recently: typical large-scale social and other networks exhibit: giant component with small diameter sparsity

More information

Overlay (and P2P) Networks

Overlay (and P2P) Networks Overlay (and P2P) Networks Part II Recap (Small World, Erdös Rényi model, Duncan Watts Model) Graph Properties Scale Free Networks Preferential Attachment Evolving Copying Navigation in Small World Samu

More information

Lesson 4. Random graphs. Sergio Barbarossa. UPC - Barcelona - July 2008

Lesson 4. Random graphs. Sergio Barbarossa. UPC - Barcelona - July 2008 Lesson 4 Random graphs Sergio Barbarossa Graph models 1. Uncorrelated random graph (Erdős, Rényi) N nodes are connected through n edges which are chosen randomly from the possible configurations 2. Binomial

More information

CS-E5740. Complex Networks. Scale-free networks

CS-E5740. Complex Networks. Scale-free networks CS-E5740 Complex Networks Scale-free networks Course outline 1. Introduction (motivation, definitions, etc. ) 2. Static network models: random and small-world networks 3. Growing network models: scale-free

More information

1 More configuration model

1 More configuration model 1 More configuration model In the last lecture, we explored the definition of the configuration model, a simple method for drawing networks from the ensemble, and derived some of its mathematical properties.

More information

Erdős-Rényi Model for network formation

Erdős-Rényi Model for network formation Network Science: Erdős-Rényi Model for network formation Ozalp Babaoglu Dipartimento di Informatica Scienza e Ingegneria Università di Bologna www.cs.unibo.it/babaoglu/ Why model? Simpler representation

More information

An Evolving Network Model With Local-World Structure

An Evolving Network Model With Local-World Structure The Eighth International Symposium on Operations Research and Its Applications (ISORA 09) Zhangjiajie, China, September 20 22, 2009 Copyright 2009 ORSC & APORC, pp. 47 423 An Evolving Network odel With

More information

MIDTERM EXAMINATION Networked Life (NETS 112) November 21, 2013 Prof. Michael Kearns

MIDTERM EXAMINATION Networked Life (NETS 112) November 21, 2013 Prof. Michael Kearns MIDTERM EXAMINATION Networked Life (NETS 112) November 21, 2013 Prof. Michael Kearns This is a closed-book exam. You should have no material on your desk other than the exam itself and a pencil or pen.

More information

Complex Networks. Structure and Dynamics

Complex Networks. Structure and Dynamics Complex Networks Structure and Dynamics Ying-Cheng Lai Department of Mathematics and Statistics Department of Electrical Engineering Arizona State University Collaborators! Adilson E. Motter, now at Max-Planck

More information

(Social) Networks Analysis III. Prof. Dr. Daning Hu Department of Informatics University of Zurich

(Social) Networks Analysis III. Prof. Dr. Daning Hu Department of Informatics University of Zurich (Social) Networks Analysis III Prof. Dr. Daning Hu Department of Informatics University of Zurich Outline Network Topological Analysis Network Models Random Networks Small-World Networks Scale-Free Networks

More information

M.E.J. Newman: Models of the Small World

M.E.J. Newman: Models of the Small World A Review Adaptive Informatics Research Centre Helsinki University of Technology November 7, 2007 Vocabulary N number of nodes of the graph l average distance between nodes D diameter of the graph d is

More information

Characteristics of Preferentially Attached Network Grown from. Small World

Characteristics of Preferentially Attached Network Grown from. Small World Characteristics of Preferentially Attached Network Grown from Small World Seungyoung Lee Graduate School of Innovation and Technology Management, Korea Advanced Institute of Science and Technology, Daejeon

More information

Graph Structure Over Time

Graph Structure Over Time Graph Structure Over Time Observing how time alters the structure of the IEEE data set Priti Kumar Computer Science Rensselaer Polytechnic Institute Troy, NY Kumarp3@rpi.edu Abstract This paper examines

More information

Small-World Models and Network Growth Models. Anastassia Semjonova Roman Tekhov

Small-World Models and Network Growth Models. Anastassia Semjonova Roman Tekhov Small-World Models and Network Growth Models Anastassia Semjonova Roman Tekhov Small world 6 billion small world? 1960s Stanley Milgram Six degree of separation Small world effect Motivation Not only friends:

More information

Small World Properties Generated by a New Algorithm Under Same Degree of All Nodes

Small World Properties Generated by a New Algorithm Under Same Degree of All Nodes Commun. Theor. Phys. (Beijing, China) 45 (2006) pp. 950 954 c International Academic Publishers Vol. 45, No. 5, May 15, 2006 Small World Properties Generated by a New Algorithm Under Same Degree of All

More information

Wednesday, March 8, Complex Networks. Presenter: Jirakhom Ruttanavakul. CS 790R, University of Nevada, Reno

Wednesday, March 8, Complex Networks. Presenter: Jirakhom Ruttanavakul. CS 790R, University of Nevada, Reno Wednesday, March 8, 2006 Complex Networks Presenter: Jirakhom Ruttanavakul CS 790R, University of Nevada, Reno Presented Papers Emergence of scaling in random networks, Barabási & Bonabeau (2003) Scale-free

More information

arxiv:cs/ v1 [cs.ir] 26 Apr 2002

arxiv:cs/ v1 [cs.ir] 26 Apr 2002 Navigating the Small World Web by Textual Cues arxiv:cs/0204054v1 [cs.ir] 26 Apr 2002 Filippo Menczer Department of Management Sciences The University of Iowa Iowa City, IA 52242 Phone: (319) 335-0884

More information

Summary: What We Have Learned So Far

Summary: What We Have Learned So Far Summary: What We Have Learned So Far small-world phenomenon Real-world networks: { Short path lengths High clustering Broad degree distributions, often power laws P (k) k γ Erdös-Renyi model: Short path

More information

Exercise set #2 (29 pts)

Exercise set #2 (29 pts) (29 pts) The deadline for handing in your solutions is Nov 16th 2015 07:00. Return your solutions (one.pdf le and one.zip le containing Python code) via e- mail to Becs-114.4150@aalto.fi. Additionally,

More information

1 Homophily and assortative mixing

1 Homophily and assortative mixing 1 Homophily and assortative mixing Networks, and particularly social networks, often exhibit a property called homophily or assortative mixing, which simply means that the attributes of vertices correlate

More information

The missing links in the BGP-based AS connectivity maps

The missing links in the BGP-based AS connectivity maps The missing links in the BGP-based AS connectivity maps Zhou, S; Mondragon, RJ http://arxiv.org/abs/cs/0303028 For additional information about this publication click this link. http://qmro.qmul.ac.uk/xmlui/handle/123456789/13070

More information

The Directed Closure Process in Hybrid Social-Information Networks, with an Analysis of Link Formation on Twitter

The Directed Closure Process in Hybrid Social-Information Networks, with an Analysis of Link Formation on Twitter Proceedings of the Fourth International AAAI Conference on Weblogs and Social Media The Directed Closure Process in Hybrid Social-Information Networks, with an Analysis of Link Formation on Twitter Daniel

More information

A Generating Function Approach to Analyze Random Graphs

A Generating Function Approach to Analyze Random Graphs A Generating Function Approach to Analyze Random Graphs Presented by - Vilas Veeraraghavan Advisor - Dr. Steven Weber Department of Electrical and Computer Engineering Drexel University April 8, 2005 Presentation

More information

Preliminary results from an agent-based adaptation of friendship games

Preliminary results from an agent-based adaptation of friendship games Preliminary results from an agent-based adaptation of friendship games David S. Dixon June 29, 2011 This paper presents agent-based model (ABM) equivalents of friendshipbased games and compares the results

More information

Social and Technological Network Data Analytics. Lecture 5: Structure of the Web, Search and Power Laws. Prof Cecilia Mascolo

Social and Technological Network Data Analytics. Lecture 5: Structure of the Web, Search and Power Laws. Prof Cecilia Mascolo Social and Technological Network Data Analytics Lecture 5: Structure of the Web, Search and Power Laws Prof Cecilia Mascolo In This Lecture We describe power law networks and their properties and show

More information

Chapter 1. Social Media and Social Computing. October 2012 Youn-Hee Han

Chapter 1. Social Media and Social Computing. October 2012 Youn-Hee Han Chapter 1. Social Media and Social Computing October 2012 Youn-Hee Han http://link.koreatech.ac.kr 1.1 Social Media A rapid development and change of the Web and the Internet Participatory web application

More information

CSE 258 Lecture 12. Web Mining and Recommender Systems. Social networks

CSE 258 Lecture 12. Web Mining and Recommender Systems. Social networks CSE 258 Lecture 12 Web Mining and Recommender Systems Social networks Social networks We ve already seen networks (a little bit) in week 3 i.e., we ve studied inference problems defined on graphs, and

More information

TELCOM2125: Network Science and Analysis

TELCOM2125: Network Science and Analysis School of Information Sciences University of Pittsburgh TELCOM2125: Network Science and Analysis Konstantinos Pelechrinis Spring 2015 2 Part 4: Dividing Networks into Clusters The problem l Graph partitioning

More information

Impact of Clustering on Epidemics in Random Networks

Impact of Clustering on Epidemics in Random Networks Impact of Clustering on Epidemics in Random Networks Joint work with Marc Lelarge INRIA-ENS 8 March 2012 Coupechoux - Lelarge (INRIA-ENS) Epidemics in Random Networks 8 March 2012 1 / 19 Outline 1 Introduction

More information

ECS 289 / MAE 298, Lecture 15 Mar 2, Diffusion, Cascades and Influence, Part II

ECS 289 / MAE 298, Lecture 15 Mar 2, Diffusion, Cascades and Influence, Part II ECS 289 / MAE 298, Lecture 15 Mar 2, 2011 Diffusion, Cascades and Influence, Part II Diffusion and cascades in networks (Nodes in one of two states) Viruses (human and computer) contact processes epidemic

More information

TELCOM2125: Network Science and Analysis

TELCOM2125: Network Science and Analysis School of Information Sciences University of Pittsburgh TELCOM2125: Network Science and Analysis Konstantinos Pelechrinis Spring 2015 Figures are taken from: M.E.J. Newman, Networks: An Introduction 2

More information

Properties of Biological Networks

Properties of Biological Networks Properties of Biological Networks presented by: Ola Hamud June 12, 2013 Supervisor: Prof. Ron Pinter Based on: NETWORK BIOLOGY: UNDERSTANDING THE CELL S FUNCTIONAL ORGANIZATION By Albert-László Barabási

More information

Failure in Complex Social Networks

Failure in Complex Social Networks Journal of Mathematical Sociology, 33:64 68, 2009 Copyright # Taylor & Francis Group, LLC ISSN: 0022-250X print/1545-5874 online DOI: 10.1080/00222500802536988 Failure in Complex Social Networks Damon

More information

CAIM: Cerca i Anàlisi d Informació Massiva

CAIM: Cerca i Anàlisi d Informació Massiva 1 / 72 CAIM: Cerca i Anàlisi d Informació Massiva FIB, Grau en Enginyeria Informàtica Slides by Marta Arias, José Balcázar, Ricard Gavaldá Department of Computer Science, UPC Fall 2016 http://www.cs.upc.edu/~caim

More information

How Do Real Networks Look? Networked Life NETS 112 Fall 2014 Prof. Michael Kearns

How Do Real Networks Look? Networked Life NETS 112 Fall 2014 Prof. Michael Kearns How Do Real Networks Look? Networked Life NETS 112 Fall 2014 Prof. Michael Kearns Roadmap Next several lectures: universal structural properties of networks Each large-scale network is unique microscopically,

More information

Social, Information, and Routing Networks: Models, Algorithms, and Strategic Behavior

Social, Information, and Routing Networks: Models, Algorithms, and Strategic Behavior Social, Information, and Routing Networks: Models, Algorithms, and Strategic Behavior Who? Prof. Aris Anagnostopoulos Prof. Luciana S. Buriol Prof. Guido Schäfer What will We Cover? Topics: Network properties

More information

CSE 190 Lecture 16. Data Mining and Predictive Analytics. Small-world phenomena

CSE 190 Lecture 16. Data Mining and Predictive Analytics. Small-world phenomena CSE 190 Lecture 16 Data Mining and Predictive Analytics Small-world phenomena Another famous study Stanley Milgram wanted to test the (already popular) hypothesis that people in social networks are separated

More information

Higher order clustering coecients in Barabasi Albert networks

Higher order clustering coecients in Barabasi Albert networks Physica A 316 (2002) 688 694 www.elsevier.com/locate/physa Higher order clustering coecients in Barabasi Albert networks Agata Fronczak, Janusz A. Ho lyst, Maciej Jedynak, Julian Sienkiewicz Faculty of

More information

Math 443/543 Graph Theory Notes 10: Small world phenomenon and decentralized search

Math 443/543 Graph Theory Notes 10: Small world phenomenon and decentralized search Math 443/543 Graph Theory Notes 0: Small world phenomenon and decentralized search David Glickenstein November 0, 008 Small world phenomenon The small world phenomenon is the principle that all people

More information

Network Theory: Social, Mythological and Fictional Networks. Math 485, Spring 2018 (Midterm Report) Christina West, Taylor Martins, Yihe Hao

Network Theory: Social, Mythological and Fictional Networks. Math 485, Spring 2018 (Midterm Report) Christina West, Taylor Martins, Yihe Hao Network Theory: Social, Mythological and Fictional Networks Math 485, Spring 2018 (Midterm Report) Christina West, Taylor Martins, Yihe Hao Abstract: Comparative mythology is a largely qualitative and

More information

ECE 158A - Data Networks

ECE 158A - Data Networks ECE 158A - Data Networks Homework 2 - due Tuesday Nov 5 in class Problem 1 - Clustering coefficient and diameter In this problem, we will compute the diameter and the clustering coefficient of a set of

More information

Attack Vulnerability of Network with Duplication-Divergence Mechanism

Attack Vulnerability of Network with Duplication-Divergence Mechanism Commun. Theor. Phys. (Beijing, China) 48 (2007) pp. 754 758 c International Academic Publishers Vol. 48, No. 4, October 5, 2007 Attack Vulnerability of Network with Duplication-Divergence Mechanism WANG

More information

CSCI5070 Advanced Topics in Social Computing

CSCI5070 Advanced Topics in Social Computing CSCI5070 Advanced Topics in Social Computing Irwin King The Chinese University of Hong Kong king@cse.cuhk.edu.hk!! 2012 All Rights Reserved. Outline Graphs Origins Definition Spectral Properties Type of

More information

CSE 158 Lecture 11. Web Mining and Recommender Systems. Social networks

CSE 158 Lecture 11. Web Mining and Recommender Systems. Social networks CSE 158 Lecture 11 Web Mining and Recommender Systems Social networks Assignment 1 Due 5pm next Monday! (Kaggle shows UTC time, but the due date is 5pm, Monday, PST) Assignment 1 Assignment 1 Social networks

More information

Models for the growth of the Web

Models for the growth of the Web Models for the growth of the Web Chi Bong Ho Introduction Yihao Ben Pu December 6, 2007 Alexander Tsiatas There has been much work done in recent years about the structure of the Web and other large information

More information

An Empirical Analysis of Communities in Real-World Networks

An Empirical Analysis of Communities in Real-World Networks An Empirical Analysis of Communities in Real-World Networks Chuan Sheng Foo Computer Science Department Stanford University csfoo@cs.stanford.edu ABSTRACT Little work has been done on the characterization

More information

Introduction to network metrics

Introduction to network metrics Universitat Politècnica de Catalunya Version 0.5 Complex and Social Networks (2018-2019) Master in Innovation and Research in Informatics (MIRI) Instructors Argimiro Arratia, argimiro@cs.upc.edu, http://www.cs.upc.edu/~argimiro/

More information

Hashing. Hashing Procedures

Hashing. Hashing Procedures Hashing Hashing Procedures Let us denote the set of all possible key values (i.e., the universe of keys) used in a dictionary application by U. Suppose an application requires a dictionary in which elements

More information

CS249: SPECIAL TOPICS MINING INFORMATION/SOCIAL NETWORKS

CS249: SPECIAL TOPICS MINING INFORMATION/SOCIAL NETWORKS CS249: SPECIAL TOPICS MINING INFORMATION/SOCIAL NETWORKS Overview of Networks Instructor: Yizhou Sun yzsun@cs.ucla.edu January 10, 2017 Overview of Information Network Analysis Network Representation Network

More information

Adaptive Robotics - Final Report Extending Q-Learning to Infinite Spaces

Adaptive Robotics - Final Report Extending Q-Learning to Infinite Spaces Adaptive Robotics - Final Report Extending Q-Learning to Infinite Spaces Eric Christiansen Michael Gorbach May 13, 2008 Abstract One of the drawbacks of standard reinforcement learning techniques is that

More information

Mathematics of networks. Artem S. Novozhilov

Mathematics of networks. Artem S. Novozhilov Mathematics of networks Artem S. Novozhilov August 29, 2013 A disclaimer: While preparing these lecture notes, I am using a lot of different sources for inspiration, which I usually do not cite in the

More information

6.207/14.15: Networks Lecture 5: Generalized Random Graphs and Small-World Model

6.207/14.15: Networks Lecture 5: Generalized Random Graphs and Small-World Model 6.207/14.15: Networks Lecture 5: Generalized Random Graphs and Small-World Model Daron Acemoglu and Asu Ozdaglar MIT September 23, 2009 1 Outline Generalized random graph models Graphs with prescribed

More information

Gaussian and Exponential Architectures in Small-World Associative Memories

Gaussian and Exponential Architectures in Small-World Associative Memories and Architectures in Small-World Associative Memories Lee Calcraft, Rod Adams and Neil Davey School of Computer Science, University of Hertfordshire College Lane, Hatfield, Herts AL1 9AB, U.K. {L.Calcraft,

More information

6. Overview. L3S Research Center, University of Hannover. 6.1 Section Motivation. Investigation of structural aspects of peer-to-peer networks

6. Overview. L3S Research Center, University of Hannover. 6.1 Section Motivation. Investigation of structural aspects of peer-to-peer networks , University of Hannover Random Graphs, Small-Worlds, and Scale-Free Networks Wolf-Tilo Balke and Wolf Siberski 05.12.07 * Original slides provided by K.A. Lehmann (University Tübingen, Germany) 6. Overview

More information

Example for calculation of clustering coefficient Node N 1 has 8 neighbors (red arrows) There are 12 connectivities among neighbors (blue arrows)

Example for calculation of clustering coefficient Node N 1 has 8 neighbors (red arrows) There are 12 connectivities among neighbors (blue arrows) Example for calculation of clustering coefficient Node N 1 has 8 neighbors (red arrows) There are 12 connectivities among neighbors (blue arrows) Average clustering coefficient of a graph Overall measure

More information

CS224W: Analysis of Networks Jure Leskovec, Stanford University

CS224W: Analysis of Networks Jure Leskovec, Stanford University CS224W: Analysis of Networks Jure Leskovec, Stanford University http://cs224w.stanford.edu 11/13/17 Jure Leskovec, Stanford CS224W: Analysis of Networks, http://cs224w.stanford.edu 2 Observations Models

More information

MAE 298, Lecture 9 April 30, Web search and decentralized search on small-worlds

MAE 298, Lecture 9 April 30, Web search and decentralized search on small-worlds MAE 298, Lecture 9 April 30, 2007 Web search and decentralized search on small-worlds Search for information Assume some resource of interest is stored at the vertices of a network: Web pages Files in

More information

On Reshaping of Clustering Coefficients in Degreebased Topology Generators

On Reshaping of Clustering Coefficients in Degreebased Topology Generators On Reshaping of Clustering Coefficients in Degreebased Topology Generators Xiafeng Li, Derek Leonard, and Dmitri Loguinov Texas A&M University Presented by Derek Leonard Agenda Motivation Statement of

More information

Dynamic network generative model

Dynamic network generative model Dynamic network generative model Habiba, Chayant Tantipathanananandh, Tanya Berger-Wolf University of Illinois at Chicago. In this work we present a statistical model for generating realistic dynamic networks

More information

CS224W Project Write-up Static Crawling on Social Graph Chantat Eksombatchai Norases Vesdapunt Phumchanit Watanaprakornkul

CS224W Project Write-up Static Crawling on Social Graph Chantat Eksombatchai Norases Vesdapunt Phumchanit Watanaprakornkul 1 CS224W Project Write-up Static Crawling on Social Graph Chantat Eksombatchai Norases Vesdapunt Phumchanit Watanaprakornkul Introduction Our problem is crawling a static social graph (snapshot). Given

More information

An Agent-Based Adaptation of Friendship Games: Observations on Network Topologies

An Agent-Based Adaptation of Friendship Games: Observations on Network Topologies An Agent-Based Adaptation of Friendship Games: Observations on Network Topologies David S. Dixon University of New Mexico, Albuquerque NM 87131, USA Abstract. A friendship game in game theory is a network

More information

Network Environments in AnyLogic. Nathaniel Osgood Agent-Based Modeling Bootcamp for Health Researchers

Network Environments in AnyLogic. Nathaniel Osgood Agent-Based Modeling Bootcamp for Health Researchers Network Environments in AnyLogic Nathaniel Osgood Agent-Based Modeling Bootcamp for Health Researchers August 23, 2011 Hands on Model Use Ahead Load model: Network Modification of SIR AB The Environment

More information

CS224W Final Report Emergence of Global Status Hierarchy in Social Networks

CS224W Final Report Emergence of Global Status Hierarchy in Social Networks CS224W Final Report Emergence of Global Status Hierarchy in Social Networks Group 0: Yue Chen, Jia Ji, Yizheng Liao December 0, 202 Introduction Social network analysis provides insights into a wide range

More information

γ : constant Goett 2 P(k) = k γ k : degree

γ : constant Goett 2 P(k) = k γ k : degree Goett 1 Jeffrey Goett Final Research Paper, Fall 2003 Professor Madey 19 December 2003 Abstract: Recent observations by physicists have lead to new theories about the mechanisms controlling the growth

More information

Response Network Emerging from Simple Perturbation

Response Network Emerging from Simple Perturbation Journal of the Korean Physical Society, Vol 44, No 3, March 2004, pp 628 632 Response Network Emerging from Simple Perturbation S-W Son, D-H Kim, Y-Y Ahn and H Jeong Department of Physics, Korea Advanced

More information

Scale Free Network Growth By Ranking. Santo Fortunato, Alessandro Flammini, and Filippo Menczer

Scale Free Network Growth By Ranking. Santo Fortunato, Alessandro Flammini, and Filippo Menczer Scale Free Network Growth By Ranking Santo Fortunato, Alessandro Flammini, and Filippo Menczer Motivation Network growth is usually explained through mechanisms that rely on node prestige measures, such

More information

E6885 Network Science Lecture 5: Network Estimation and Modeling

E6885 Network Science Lecture 5: Network Estimation and Modeling E 6885 Topics in Signal Processing -- Network Science E6885 Network Science Lecture 5: Network Estimation and Modeling Ching-Yung Lin, Dept. of Electrical Engineering, Columbia University October 7th,

More information

TELCOM2125: Network Science and Analysis

TELCOM2125: Network Science and Analysis School of Information Sciences University of Pittsburgh TELCOM2125: Network Science and Analysis Konstantinos Pelechrinis Spring 2015 Figures are taken from: M.E.J. Newman, Networks: An Introduction 2

More information

Spatial Patterns Point Pattern Analysis Geographic Patterns in Areal Data

Spatial Patterns Point Pattern Analysis Geographic Patterns in Areal Data Spatial Patterns We will examine methods that are used to analyze patterns in two sorts of spatial data: Point Pattern Analysis - These methods concern themselves with the location information associated

More information

Small-world networks

Small-world networks Small-world networks c A. J. Ganesh, University of Bristol, 2015 Popular folklore asserts that any two people in the world are linked through a chain of no more than six mutual acquaintances, as encapsulated

More information

DOWNLOAD PDF BIG IDEAS MATH VERTICAL SHRINK OF A PARABOLA

DOWNLOAD PDF BIG IDEAS MATH VERTICAL SHRINK OF A PARABOLA Chapter 1 : BioMath: Transformation of Graphs Use the results in part (a) to identify the vertex of the parabola. c. Find a vertical line on your graph paper so that when you fold the paper, the left portion

More information

arxiv: v1 [cs.si] 11 Jan 2019

arxiv: v1 [cs.si] 11 Jan 2019 A Community-aware Network Growth Model for Synthetic Social Network Generation Furkan Gursoy furkan.gursoy@boun.edu.tr Bertan Badur bertan.badur@boun.edu.tr arxiv:1901.03629v1 [cs.si] 11 Jan 2019 Dept.

More information

Solving Traveling Salesman Problem Using Parallel Genetic. Algorithm and Simulated Annealing

Solving Traveling Salesman Problem Using Parallel Genetic. Algorithm and Simulated Annealing Solving Traveling Salesman Problem Using Parallel Genetic Algorithm and Simulated Annealing Fan Yang May 18, 2010 Abstract The traveling salesman problem (TSP) is to find a tour of a given number of cities

More information

Behavioral Data Mining. Lecture 9 Modeling People

Behavioral Data Mining. Lecture 9 Modeling People Behavioral Data Mining Lecture 9 Modeling People Outline Power Laws Big-5 Personality Factors Social Network Structure Power Laws Y-axis = frequency of word, X-axis = rank in decreasing order Power Laws

More information

Lecture (08, 09) Routing in Switched Networks

Lecture (08, 09) Routing in Switched Networks Agenda Lecture (08, 09) Routing in Switched Networks Dr. Ahmed ElShafee Routing protocols Fixed Flooding Random Adaptive ARPANET Routing Strategies ١ Dr. Ahmed ElShafee, ACU Fall 2011, Networks I ٢ Dr.

More information

CSE 158 Lecture 13. Web Mining and Recommender Systems. Triadic closure; strong & weak ties

CSE 158 Lecture 13. Web Mining and Recommender Systems. Triadic closure; strong & weak ties CSE 158 Lecture 13 Web Mining and Recommender Systems Triadic closure; strong & weak ties Monday Random models of networks: Erdos Renyi random graphs (picture from Wikipedia http://en.wikipedia.org/wiki/erd%c5%91s%e2%80%93r%c3%a9nyi_model)

More information

Topology Enhancement in Wireless Multihop Networks: A Top-down Approach

Topology Enhancement in Wireless Multihop Networks: A Top-down Approach Topology Enhancement in Wireless Multihop Networks: A Top-down Approach Symeon Papavassiliou (joint work with Eleni Stai and Vasileios Karyotis) National Technical University of Athens (NTUA) School of

More information

Chapter 10. Fundamental Network Algorithms. M. E. J. Newman. May 6, M. E. J. Newman Chapter 10 May 6, / 33

Chapter 10. Fundamental Network Algorithms. M. E. J. Newman. May 6, M. E. J. Newman Chapter 10 May 6, / 33 Chapter 10 Fundamental Network Algorithms M. E. J. Newman May 6, 2015 M. E. J. Newman Chapter 10 May 6, 2015 1 / 33 Table of Contents 1 Algorithms for Degrees and Degree Distributions Degree-Degree Correlation

More information

CSE 158 Lecture 11. Web Mining and Recommender Systems. Triadic closure; strong & weak ties

CSE 158 Lecture 11. Web Mining and Recommender Systems. Triadic closure; strong & weak ties CSE 158 Lecture 11 Web Mining and Recommender Systems Triadic closure; strong & weak ties Triangles So far we ve seen (a little about) how networks can be characterized by their connectivity patterns What

More information

Supplementary material to Epidemic spreading on complex networks with community structures

Supplementary material to Epidemic spreading on complex networks with community structures Supplementary material to Epidemic spreading on complex networks with community structures Clara Stegehuis, Remco van der Hofstad, Johan S. H. van Leeuwaarden Supplementary otes Supplementary ote etwork

More information

Critical Phenomena in Complex Networks

Critical Phenomena in Complex Networks Critical Phenomena in Complex Networks Term essay for Physics 563: Phase Transitions and the Renormalization Group University of Illinois at Urbana-Champaign Vikyath Deviprasad Rao 11 May 2012 Abstract

More information

CSE 255 Lecture 13. Data Mining and Predictive Analytics. Triadic closure; strong & weak ties

CSE 255 Lecture 13. Data Mining and Predictive Analytics. Triadic closure; strong & weak ties CSE 255 Lecture 13 Data Mining and Predictive Analytics Triadic closure; strong & weak ties Monday Random models of networks: Erdos Renyi random graphs (picture from Wikipedia http://en.wikipedia.org/wiki/erd%c5%91s%e2%80%93r%c3%a9nyi_model)

More information

Volume 2, Issue 11, November 2014 International Journal of Advance Research in Computer Science and Management Studies

Volume 2, Issue 11, November 2014 International Journal of Advance Research in Computer Science and Management Studies Volume 2, Issue 11, November 2014 International Journal of Advance Research in Computer Science and Management Studies Research Article / Survey Paper / Case Study Available online at: www.ijarcsms.com

More information

Link Prediction for Social Network

Link Prediction for Social Network Link Prediction for Social Network Ning Lin Computer Science and Engineering University of California, San Diego Email: nil016@eng.ucsd.edu Abstract Friendship recommendation has become an important issue

More information

CS224W: Social and Information Network Analysis Jure Leskovec, Stanford University

CS224W: Social and Information Network Analysis Jure Leskovec, Stanford University CS224W: Social and Information Network Analysis Jure Leskovec, Stanford University http://cs224w.stanford.edu 10/4/2011 Jure Leskovec, Stanford CS224W: Social and Information Network Analysis, http://cs224w.stanford.edu

More information

Intro to Random Graphs and Exponential Random Graph Models

Intro to Random Graphs and Exponential Random Graph Models Intro to Random Graphs and Exponential Random Graph Models Danielle Larcomb University of Denver Danielle Larcomb Random Graphs 1/26 Necessity of Random Graphs The study of complex networks plays an increasingly

More information

THE EFFECT OF SEGREGATION IN NON- REPEATED PRISONER'S DILEMMA

THE EFFECT OF SEGREGATION IN NON- REPEATED PRISONER'S DILEMMA THE EFFECT OF SEGREGATION IN NON- REPEATED PRISONER'S DILEMMA Thomas Nordli University of South-Eastern Norway, Norway ABSTRACT This article consolidates the idea that non-random pairing can promote the

More information

The Mathematical Description of Networks

The Mathematical Description of Networks Modelling Complex Systems University of Manchester, 21 st 23 rd June 2010 Tim Evans Theoretical Physics The Mathematical Description of Networs Page 1 Notation I will focus on Simple Graphs with multiple

More information

Connections in Networks: A Hybrid Approach

Connections in Networks: A Hybrid Approach Connections in Networks: A Hybrid Approach Carla P. Gomes 1 Willem-Jan van Hoeve 2 Ashish Sabharwal 1 1 Department of Computer Science, Cornell University, Ithaca NY 14853, U.S.A. {gomes,sabhar}@cs.cornell.edu

More information

Network community detection with edge classifiers trained on LFR graphs

Network community detection with edge classifiers trained on LFR graphs Network community detection with edge classifiers trained on LFR graphs Twan van Laarhoven and Elena Marchiori Department of Computer Science, Radboud University Nijmegen, The Netherlands Abstract. Graphs

More information

Weighted Alternating Least Squares (WALS) for Movie Recommendations) Drew Hodun SCPD. Abstract

Weighted Alternating Least Squares (WALS) for Movie Recommendations) Drew Hodun SCPD. Abstract Weighted Alternating Least Squares (WALS) for Movie Recommendations) Drew Hodun SCPD Abstract There are two common main approaches to ML recommender systems, feedback-based systems and content-based systems.

More information

CS 224W Final Report Group 37

CS 224W Final Report Group 37 1 Introduction CS 224W Final Report Group 37 Aaron B. Adcock Milinda Lakkam Justin Meyer Much of the current research is being done on social networks, where the cost of an edge is almost nothing; the

More information

caution in interpreting graph-theoretic diagnostics

caution in interpreting graph-theoretic diagnostics April 17, 2013 What is a network [1, 2, 3] What is a network [1, 2, 3] What is a network [1, 2, 3] What is a network [1, 2, 3] What is a network a collection of more or less identical agents or objects,

More information

Hyperbolic Geometry of Complex Network Data

Hyperbolic Geometry of Complex Network Data Hyperbolic Geometry of Complex Network Data Konstantin Zuev http://www.its.caltech.edu/~zuev/ Joint work with D. Krioukov, M. Boguñá, and G. Bianconi CMX seminar, Caltech May 24, 2017 How do complex networks

More information

CS224W: Social and Information Network Analysis Jure Leskovec, Stanford University, y http://cs224w.stanford.edu Due in 1 week: Oct 4 in class! The idea of the reaction papers is: To familiarize yourselves

More information

Whitepaper US SEO Ranking Factors 2012

Whitepaper US SEO Ranking Factors 2012 Whitepaper US SEO Ranking Factors 2012 Authors: Marcus Tober, Sebastian Weber Searchmetrics Inc. 1115 Broadway 12th Floor, Room 1213 New York, NY 10010 Phone: 1 866-411-9494 E-Mail: sales-us@searchmetrics.com

More information

4.12 Generalization. In back-propagation learning, as many training examples as possible are typically used.

4.12 Generalization. In back-propagation learning, as many training examples as possible are typically used. 1 4.12 Generalization In back-propagation learning, as many training examples as possible are typically used. It is hoped that the network so designed generalizes well. A network generalizes well when

More information

Graph Mining and Social Network Analysis

Graph Mining and Social Network Analysis Graph Mining and Social Network Analysis Data Mining and Text Mining (UIC 583 @ Politecnico di Milano) References q Jiawei Han and Micheline Kamber, "Data Mining: Concepts and Techniques", The Morgan Kaufmann

More information

Chapter 14 Global Search Algorithms

Chapter 14 Global Search Algorithms Chapter 14 Global Search Algorithms An Introduction to Optimization Spring, 2015 Wei-Ta Chu 1 Introduction We discuss various search methods that attempts to search throughout the entire feasible set.

More information

Two-dimensional Totalistic Code 52

Two-dimensional Totalistic Code 52 Two-dimensional Totalistic Code 52 Todd Rowland Senior Research Associate, Wolfram Research, Inc. 100 Trade Center Drive, Champaign, IL The totalistic two-dimensional cellular automaton code 52 is capable

More information