Trends Prediction Using Social Diffusion Models

The importance of the ability of predict trends in social media has been growing rapidly in the past few years with the growing dominance of social media in our everyday’s life. Whereas many works focus on the detection of anomalies in networks, there exist little theoretical work on the prediction of the likelihood of anomalous network pattern to globally spread and become “trends”. In this work we present an analytic model the social diffusion dynamics of spreading network patterns. Our proposed method is based on information diffusion models, and is capable of predicting future trends based on the analysis of past social interactions between the community’s members. We present an analytic lower bound for the probability that emerging trends would successful spread through the network. We demonstrate our model using two comprehensive social datasets - the “Friends and Family” experiment that was held in MIT for over a year, where the complete activity of 140 users was analyzed, and a financial dataset containing the complete activities of over 1.5 million members of the “eToro” social trading community.

💡 Research Summary

The paper tackles the problem of predicting whether an observed anomalous pattern in a social network will evolve into a widespread trend. While much prior work focuses on anomaly detection, the authors aim to provide a theoretical framework that quantifies the probability of global diffusion given a snapshot of the network’s current adoption state.

The authors model the social network as a scale‑free graph G(n, c, γ) where node degrees follow a power‑law distribution. Each user exposed to a potential trend generates, on average, β “exposure agents” that perform random‑walk‑like transmissions along social links. This formulation captures the transitive nature of exposure: an exposed user creates agents that can further expose their neighbors, and so on.

Key variables are defined: Vₐ(t) denotes the set of users advocating the trend at time t; β is the diffusion factor; P_Δ is the probability that two random nodes differ in degree by at least a factor Δ; σ₋ is a “low temporal resistance” term that aggregates β, P_Δ and the time interval Δt; P_Local‑Adopt(v, t, Δt) is the individual adoption probability, modeled as 1 − exp