Social Ties and Information Diffusion Kristina Lerman University of Southern California CS 599: Social Media Analysis University of Southern California 1 Strength of social ties and information diffusion link = social tie Not all ties are created equal How do micro-level structure (e.g., strength of social ties) affect macro-level phenomena (e.g., information diffusion) Strength of social ties: three views • Theoretical analysis – Granovetter, M. (1973) “The Strong of Weak Ties” • Empirical analysis (mobile phone data) – Onnela et al (2007) “Structure and ties strength in mobile communication networks” • Experimental study (Facebook) – Bakshy et al (2012) “The Role of Social Networks in Information Diffusion” Social ties • Strength of a tie between two people is characterized by – Frequency and length of interactions – Emotional intensity of interactions – Intimacy, … • Strong ties vs weak ties Strong ties • Friends • Family Weak ties • Acquaintances • Co-workers • Neighbors Measuring tie strength • Intensity and duration of social interactions must be reflected in network structure • Consider nodes A and B – Set S={C, D, E, …} of all nodes tied to either or both A & B • Hypothesis: the stronger the tie between A & B, the larger the number of nodes in S to whom both A & B are tied C D E F C A A D B B E F Measuring tie strength • Strong ties: large overlap in friendship circles – These people spend much time together • Weak ties: small overlap in friendship circles – These people occasionally spend time together • No ties: no overlap in friendship circles C D E F C A A D B B E F A “bridge” • Bridge – a line in a network which provides the only path between two nodes, e.g., A-B – Link A-B provides the only route for information and influence to flow between communities • All bridges are weak ties – Weak ties play an important role in diffusion A B A “bridge” • Bridge – a line in a network which provides the only path between two nodes, e.g., A-B – Link A-B provides the only route for information and influence to flow between communities • All bridges are weak ties – Weak ties play an important role in diffusion A B Weak ties in diffusion processes • Diffusing item can reach more nodes and traverse greater social distance, when passing over weak ties rather than strong ties – If it does not pass over weak ties (bridges), diffusion will be localized to cliques • Surprisingly, no direct empirical evidence from sociology literature (circa 1973), but indirect evidence A B Indirect evidence from sociological studies • In a variation of Milgram’s “small world” experiment, white “senders” were asked to forward a letter to a black “target” – Critical link is the first time the letter crosses from white to black communities – 50% of links where the white described the black as an acquaintance (weak tie) were successful – 26% of links where the white described the black as a friend (strong tie) were successful weak ties are more effective in bridging social distance • Study of middle school students [Rapoport & Horvath 1961] – Students asked to nominate 1st, 2nd, etc best friend Many more students are reached through weak ties (7th and 8th best choices) than strong ties (1st and 2nd choices) Job search study • Workers find out about new jobs through personal contacts • How does worker-contact tie strength affect chances of learning about the new job? – Measure tie strength by frequency of contact • Often (at least once a week) • Occasionally (less than 2x/week, more than 1x/year) • Rarely (1x/year or less) What kind of contacts helped workers (N=54) find jobs? Job search study • Chance encounters with weak ties (college friends, former workmates) provided most of the job leads – “It is remarkable that people receive crucial information form individuals whose very existence they have forgotten” What kind of contacts helped workers (N=54) find jobs? Length of chains • Granovetter traced job lead back to its source, i.e., where did the contact get information about the job from? – direct lead from the employer length 1 – Lead from a single intermediary length 2 Similar to getting job lead from ad in newspaper Tie strength and community organizing • “Why do some communities organize for common goals easily and effectively whereas others seem unable to mobilize resources, even against dire threats?” – In the absence of weak ties, community is partitioned in non-interacting cliques Cannot organize Can organize Summary • Personal experience depends on large-scale aspects of social structure • Weak ties necessary for receiving novel information and opportunities • Strong ties lead to social cohesion, but overall fragmentation • Open questions – What about tie content, not just strength? – How to handle negative ties? – More empirical evidence? Structure and Tie Strengths in Mobile Communication Networks J.P. Onnela, J. Saramaki, J. Hyvonen, G. Szabo, D. Lazer, K. Kaski, J. Kertesz, and A.L. Barabasi Presented by Arul Samuel Rajkumar Focus of the Paper • Trends in Mobile Communication Networks – Mobile Calls • Understand the Tie Strengths – Weak and Strong Ties • Understand the Network topology in Mobiles – Community structure and its links • Role of Tie Strengths and Network Topology in Information Diffusion Concepts • Tie Strengths weights on the edges / links Assumption: they affect the information diffusion rate. • Weights are Call duration • Stronger Tie - more information is shared. Concepts • Network Topology Structure of the network – Position of strong ties and weak ties in the network. Weak ties are bridges between communities • Information Diffusion Rate The amount of information flowing through the links at a specified duration. Data Collection • Mobile Network Calls 18 months calls from an anonymous country Assumption: Calls ~ Other mediums of communication in Mobile communications • Represented as Dyadic connections Properties : – One to One connection – Intentional Sharing or spread of information between individuals– not broadcast • Referred to as MCG or Mobile Call Graph Distributions of MCG Link weight distributio n Degree Distribut ion • Fat tailed (skewed Power law) – Majority connected with ~10 people (friends circle) – Few people with dozens of people • Fat tailed • few people talk for long time • Majority of people talk for short duration Network Topology in MCG A – communities by link strength B – Randomly permuted link strength C – Link strength by betweenness centrality Topological Overlap 0 - no overlap 1 - complete overlap Network disintegration on Node removal black - removal of ties with high weights (A) | high <O> (B) red - removal of ties with low weights (A) | high <O> (B) The spike occurs when the frequency f become the critical frequency fc where Real vs Control Simulated networks How information diffusion takes place? • Rate of Spreading • Community trapping • Average weighted ties Who gets the information? • Intermediate Tie Strength Information Diffusion Trends •Real Communication Networks Inactive parts due to the weak links (slower information transfer) •Controlled Communication Networks Information spreads fast due to the average weights on the links Conclusions • Correlation Trends - Network topology and tie strengths are correlated even in a dyadic network like the mobile communication. • Network disintegration - Removal of weak ties disintegrate the network, but removal of strong ties only reduce the scale of the network. • Network topology affects the rate of information diffusion • Role of strong & weak ties is limited in MCF as intermediate nodes get the information. Limitations & New Ideas • Mobile calls only exhibit limited communication patterns • Emails / Messages provides additional connections patterns. • Relevance of topic is also taken into consideration for individuals to spread information. • Relevance of Group for the individual to spread or share information. Key Take-aways • Modelling network links using Tie strengths. (weights) • Analyzing the effect of removal of network ties. (links) • Controlling rate of Information diffusion by tie strengths and network topology Questions The Role of Social Networks in Information Diffusion Eytan Bakshy, Cameron Marlow, Itamar Rosenn, Lada Adamic Presented by Manas Jog Problem Introduction Objective 1. Identify the importance of social network in enabling information propagation. 2. Role of strong and weak ties in diffusion. Challenges Separate influence from: Homophily Tendency of individuals with similar characteristics to associate with one another. Result: Similar information sources External correlations External influence via email, IM or other socia networking sites. Result: Internal & external influences mixed! Approach Main idea For each URL: Randomly assign every user into two conditions: No feed condition: User must not see the URL their friends shared. Feed condition: User can see the URL their friends shared. Allows us to differentiate internal influence from external influence. Causal relationship between Facebook Feed and external influences can now be blocked. Collecting Data 253,238,367 users 75,888,466 URLs 7 weeks Temporal Clustering Analysis Find behavior changes after an individual shares a link Subjects share link soon after their friends share, even when shares are hidden, but only slightly later Effect of online contagion Tie Strength and Influence Measuring Tie Strength Tie Strength is directly proportional to the number of interactions. Tie Strength and Influence Inference 1. Subject shares more when tie strength is strong. 2. Multiplicative effect of feed diminishes with tie strength. Similar trends for other interactions Inferences 1. Tie strength is stronger predictor of external influence. 2. Weak ties carry novel information that a user is unlikely to be exposed to. 3. Weak ties increase diversity of information. Aggregating tie impact Assumption One or more interactions is considered as strong tie. Strong ties individually more influential, but weak ties collectively far more influential Thank you! Questions? Manas Jog Summary • Strong ties – surrounded by many mutual friends – characterized by lots of shared time together • Weak ties – have few mutual friends – Serve as bridges to diverse parts of the network – Provide access to novel information