Lev Muchnik

Identification of influential spreaders in complex networks

Spreading of information, ideas or diseases can be conveniently modelled in the context of complex networks. An analysis now reveals that the most efficient spreaders are not always necessarily the most connected agents in a network. Instead, the position of an agent relative to the hierarchical topological organization of the network might be as important as its connectivity.

Social Influence Bias: A Randomized Experiment

Follow the Leader? The Internet has increased the likelihood that our decisions will be influenced by those being made around us. On the one hand, group decision-making can lead to better decisions, but it can also lead to “herding effects” that have resulted in financial disasters. Muchnik et al. (p. 647) examined the effect of collective information via a randomized experiment, which involved collaboration with a social news aggregation Web site on which readers could vote and comment on posted comments. Data were collected and analyzed after the Web site administrators arbitrarily voted positively or negatively (or not at all) as the first comment on more than 100,000 posts. False positive entries led to inflated subsequent scores, whereas false negative initial votes had small long-term effects. Both the topic being commented upon and the relationship between the poster and commenter were important. Future efforts will be needed to sort out how to correct for such effects in polls or other collective intelligence systems in order to counter social biases. A social news aggregation Web site was used to test whether prior ratings influence others to create bias in rating behavior. Our society is increasingly relying on the digitized, aggregated opinions of others to make decisions. We therefore designed and analyzed a large-scale randomized experiment on a social news aggregation Web site to investigate whether knowledge of such aggregates distorts decision-making. Prior ratings created significant bias in individual rating behavior, and positive and negative social influences created asymmetric herding effects. Whereas negative social influence inspired users to correct manipulated ratings, positive social influence increased the likelihood of positive ratings by 32% and created accumulating positive herding that increased final ratings by 25% on average. This positive herding was topic-dependent and affected by whether individuals were viewing the opinions of friends or enemies. A mixture of changing opinion and greater turnout under both manipulations together with a natural tendency to up-vote on the site combined to create the herding effects. Such findings will help interpret collective judgment accurately and avoid social influence bias in collective intelligence in the future.

Engineering Social Contagions: Optimal Network Seeding in the Presence of Homophily

We use data on a real, large-scale social network of 27 million individuals interacting daily, together with the day-by-day adoption of a new mobile service product, to inform, build, and analyze data-driven simulations of the effectiveness of seeding (network targeting) strategies under different social conditions. Three main results emerge from our simulations. First, failure to consider homophily creates significant overestimation of the effectiveness of seeding strategies, casting doubt on conclusions drawn by simulation studies that do not model homophily. Second, seeding is constrained by the small fraction of potential influencers that exist in the network. We find that seeding more than 0.2% of the population is wasteful because the gain from their adoption is lower than the gain from their natural adoption (without seeding). Third, seeding is more effective in the presence of greater social influence. Stronger peer influence creates a greater than additive effect when combined with seeding. Our findings call into question some conventional wisdom about these strategies and suggest that their overall effectiveness may be overestimated.

Exploring the Complex Pattern of Information Spreading in Online Blog Communities

Information spreading in online social communities has attracted tremendous attention due to its utmost practical values in applications. Despite that several individual-level diffusion data have been investigated, we still lack the detailed understanding of the spreading pattern of information. Here, by comparing information flows and social links in a blog community, we find that the diffusion processes are induced by three different spreading mechanisms: social spreading, self-promotion and broadcast. Although numerous previous studies have employed epidemic spreading models to simulate information diffusion, we observe that such models fail to reproduce the realistic diffusion pattern. In respect to users behaviors, strikingly, we find that most users would stick to one specific diffusion mechanism. Moreover, our observations indicate that the social spreading is not only crucial for the structure of diffusion trees, but also capable of inducing more subsequent individuals to acquire the information. Our findings suggest new directions for modeling of information diffusion in social systems, and could inform design of efficient propagation strategies based on users behaviors.

Miniature self-contained intravascular magnetic resonance (IVMI) probe for clinical applications

A miniature (1.73 mm in diameter) NMR probe, which contains a magnet and a radiofrequency (RF) coil, is presented. This probe is integrated at the tip of a standard catheter and can be inserted into the human coronary arteries, creating local magnetic fields needed to obtain the NMR signal from the blood vessel walls, without the need for external magnet or RF coils. The basic theory governing the probe performance in terms of signal‐to‐noise‐ratio and contrast parameters is presented, along with measured results from test samples. The NMR signal can be analyzed to obtain tissue contrast parameters such as T1, T2 and the diffusion coefficient, which may be used to detect lipid‐rich vulnerable plaques in the coronary arteries. Magn Reson Med 54:105–112, 2005.

Design of Randomized Experiments in Networks

Over the last decade, the emergence of pervasive online and digitally enabled environments has created a rich source of detailed data on human behavior. Yet, the promise of big data has recently come under fire for its inability to separate correlation from causation-to derive actionable insights and yield effective policies. Fortunately, the same online platforms on which we interact on a day-to-day basis permit experimentation at large scales, ushering in a new movement toward big experiments. Randomized controlled trials are the heart of the scientific method and when designed correctly provide clean causal inferences that are robust and reproducible. However, the realization that our world is highly connected and that behavioral and economic outcomes at the individual and population level depend upon this connectivity challenges the very principles of experimental design. The proper design and analysis of experiments in networks is, therefore, critically important. In this work, we categorize and review the emerging strategies to design and analyze experiments in networks and discuss their strengths and weaknesses.

Line graphs as social networks

It was demonstrated recently that the line graphs are clustered and assortative. These topological features are known to characterize some social networks [M.E.J. Newman, Y. Park, Why social networks are different from other types of networks, Phys. Rev. E 68 (2003) 036122]; it was argued that this similarity reveals their cliquey character. In the model proposed here, a social network is the line graph of an initial network of families, communities, interest groups, school classes and small companies. These groups play the role of nodes, and individuals are represented by links between these nodes. The picture is supported by the data on the LiveJournal network of about 8×106 people.

Scaling and Memory in Return Loss Intervals: Application to Risk Estimation

We study the statistics of the return intervals τ q between two consecutive return losses below a threshold −q, in various stocks, currencies and commodities. We find the probability distribution function (pdf) of τ q scales with the mean return interval τ q in a quite universal way, which may enable us to extrapolate rare events from the behavior of more frequent events with better statistics. The functional form of the pdf shows deviation from a simple exponential behavior, suggesting memory effects in losses. The memory shows up strongly in the conditional mean loss return intervals which depend significantly on the previous interval. This dependence can be used to improve the estimate of the risk level.

Identity and Opinion: A Randomized Experiment

Identity cues – pseudonyms or real names, often displayed with profile photos – appear ubiquitously alongside content in social media. In this paper, we seek to understand to what extent these cues affect how people form opinions about content they consume online. We present results from a large scale (N=1.7×10^7), two-year-long field experiment with a novel anonymization condition. We assigned content produced on a social news discussion website to “identified�? and “anonymous�? conditions to estimate the causal effect of identity cues on how viewers interact with content in terms of ratings and reply comments. Our results show that identity cues cause people to rate content faster (consistent with heuristic processing) and to rate according to con- tent producers’ production, reputation, and reciprocal ratings with content viewers. Our results provide insight into the evolution of status in online communities and evidence for rich-get-richer dynamics that are mediated by identity cues. The methods we use can help platform providers detect and correct for an important source of bias in crowdsourced ratings that cause inequality in feedback.

Agent Perception Modeling for Movement in Crowds

This paper explores the integration of a perception map to an agent based model simulated on a realistic physical space. Each agent’s perception map stores density information about the physical space which is used for routing. The scenario considered is the evacuation of a space given a crowd. Through agent interactions, both in physical proximity and through distant communications, agents update their perception maps and continuously work to overcome their incomplete perception of the world. Overall, this work aims at investigating the dynamics of agent information diffusion for emergency scenarios and combines three general elements: (1) an agent-based simulation of crowd dynamics in an emergency scenario over a real physical space, (2) a sophisticated decision making process driven by the agent’s subjective view of the world and effected by trust, belief and confidence, and (3) agent’s activity aimed at building relationships with specific peers that is based on mutual benefit from sharing information.