netsci 2010 in boston was a success! on friday at 4pm, we’ll have a discussion session in the computer lab of cambridge university (all invited! slides are here). i’ve blogged about the satellite events on monday (round table at harvard) and on tuesday (mobile network workshop). here are the notes i’ve taken during the actual conference on wednesday, thursday, and friday:
in his talk, newman (who just published a really good textbook on networks) mathematically proved that your friends have more friends than you do (if <k> is your avg degree, then your neighbors’ avg degree tend to be <k^2>/<k>). he then empirically confirmed this finding upon 3 nets – net of biologists, net of mathematicians, and the Internet. newman has been also interested in the following problems:
1) what’s the probability that a given network node belongs to a giant connected component of size S (see expression pi_S in his paper)
2) what’s the probability a paper/court case will be cited over time. the model works surprisingly well. as one expects, papers are cited a lot immediately after they are published and then citation count dies off
3) how motifs such as clustering (tendency for nets to form triangles – the friends of my friends are also my friends) and diamonds tend to be created – see 1 and 2
folks in madrid characterised time dynamics of single social links by using 3 measures: intensity (amount of time spent in the communication), stability (diff between first and last communication), and homogeneity (std deviation divided by the mean). real communication is very heterogeneous and unstable (compared to a null model where strengths are randomly shuffled). it turns out that stability is the best predictor for tie strength and intensity is the second most important predictor.
david lazer gave a fantastic talk exploring the question of how we think together and, more specifically, how to collectively solve problems. humans tend to solve problems balancing two competing desires: exploration (looking for new solutions) vs. exploitation (taking advantage of what you know). david’s research has shown that net structure affects the balance one places betw exploration and exploitation (because net structure affects dynamics of info diffusion, solution retention, and info diversity). interestingly, if a group is more connected, its members aren’t necessarily more effective in solving problems. to show this, david has run an agent-based simulation and a variety of human experiments:
- using an agent-based model, david showed that the ability of solving problems: 1) has an initial peak and then stabilises, if people are connected by a fully connected network (which apparently kills heterogeneity); 2) constantly and slowly increases, if people are connected by a linear network. those 2 results suggest that, in the short run, it’s good to have a fully connected network. however, in the long run, letting people look at each other’s performance is counterproductive.
- using human experiments in which networked people tried to solve the travels-man problem under two different treatments – a) sees none, sees all (show how your peers are doing); b) memory (show how you’ve done in the past), no memory – david found that communication reduces exploration & allows evaluation of performance (one can easily retain one’s best solution).
waber studied face-to-face contact networks in different organisations and found that few regular communication patterns tend to emerge. they call one of those patterns kith (which is similar to what burt calls ‘constraint’) and showed that kith is positively correlated with job satisfaction and, surprisingly, does not change with net size, density, and hierarchical structure . this invariance could be generated by cognitive limitations of each individual – each of us can cope with a group of limited size – see dunbar number – and has limited attention span. in the future, ben
will also look at where (in which room) people tend to gather and talk.
sinan aral (who is organising winworkshop.net) briefly presented his recent PNAS article on causality in behavioural contagion. the idea is that causualty in information spreading is difficult to assess and cannot be determined only based on longitudinal data – homophily may well explain how people become ‘infected’. to unearth casual relationships, one should use randomised assignments (see falk & heckman, science (2009)). aral and his group run a large number of randomised trials on Facebook (see paper). the goal was to study how a (commercial) application would be adopted. to that end, they assigned users to two groups: 1. an experimental group for which viral messaging is enabled (they can invite their friends to adopt the application); 2. a control group for which no viral messaging is allowed. they then observed the adoption of the application under those two conditions. they found that notifications of what your friends are doing with the applications are annoying. friends of passive users adopt 10x more.
jp onnela presented his paper on community detection in multi-slice networks (published in science this month). here are his slides.the paper presents a way of representing different types of networks (slices) regarding the same people (e.g., email network and contact network of people in the same company) and a way of identify communities across those slices. their proposal has been applied to facebook data from which different nets (facebook friends, facebook pic friends, roommates, housing group net) have been extracted. they throughly studies how the communities across the different slices/nets overlap.
in his talk, rafael brune of northwestern defined short, medium, and long geographical links and run betweenness centrality using random walks. for each node, he counted the number of times the walk passed through links of different types (e.g., short-short links, short-med links, med-long links, etc.).
ibm folks in boston (lead by kate ehrlich) are studying ways of countering excess of homophily in social-networking services. they analysed 3-month log data from a microblogging tool within IBM. the data is about 881 users who post status updates (limited to 250 characters). here are few communication network stats: density=0.6, avg distance=3.34, transitivity=5.39%, reciprocity=32%. the idea behind this work is to connect people based on 5 types of brokerage (as described in “Structures of Mediation: A Formal Approach to Brokerage in Transaction Networks”): 1) coordinator (connects within group); 2) gatekeeper (buffer for incoming); 3) representative (external communication); 4) consultant; and 5) liaison.
renaud lambiotte of imperial considered a multiplayer game and extracted different networks from it (communication net, trade net, friendship net, attack net, bounty net, enemy net). by studying those multiplex networks, he found that:
. positive net are reciprocal, show high cohesion, and aren’t power-law
. negative net are non-reciprocal, show low cohesion, and are power-law
renaud also tested the theory of social balance on positive and negative networks (see Dynamics of social balance on networks) with a null model. here are his slides.
jure leskovec presented his study of 3 networks containing both positive and negative ties with the purpose of testing a variety of social theories. i wonder whether the three datasets are fit for purpose – the context of each social tie is missing (that is, all ties are equal).
omar lizardo (with david hachen) studied the relationship between tie reciprocity and node degree. he found that degree-similar dyads tend to be reciprocal and strong, and strong ties are likely to persist over time.
gourab ghoshal co-authored with Newman a work on random hypergraphs (graphs in which nodes are of different nature) and their applications in folksnomies.
vedres & colleagues studied a network of companies and identified groups in this network (paper). they tried to explain the performance of each company as a function of the characteristics of the group(s) the company belongs to (as a function of inter-cohesion, stability, governmental tie, etc.). interestingly, they refer to the article “the persistence of social groups” published by simmel in 1898.