Source-linked AI summary
The Anatomy of a Scientific Rumor
M. De Domenico, A. Lima, P. Mougel, M. Musolesi
TL;DR
The paper examines how Higgs-discovery information spread on Twitter before, during, and after the 4 July 2012 announcement. It analyzes user activity and network structure, then models activation dynamics; the model reproduces the collective behavior of about 500,000 users with remarkable accuracy.
Problem
The paper investigates spatio-temporal information spreading and time-varying user activity during a major scientific announcement on Twitter.
Method
The study analyzes Higgs-related tweets and their follower network, defining active users by whether they tweeted about the Higgs boson at a given time and modeling activation through neighborhood social reinforcement.
Results
The model reproduces the collective behavior of about 500,000 users with remarkable accuracy, while activity shifts from localized rapid tweeting before the announcement to frenetic, spatially diffuse activity during it.
Takeaways & Limitations
The proposed framework can be applied and fitted to other information-spreading processes on online and physical social networks.
Takeaways & Limitations
The model’s governing equations generally require simplifying assumptions or numerical methods because their analytical solution is unavailable.
Abstract
from arXiv · showhide
The announcement of the discovery of a Higgs boson-like particle at CERN will be remembered as one of the milestones of the scientific endeavor of the 21st century. In this paper we present a study of information spreading processes on Twitter before, during and after the announcement of the discovery of a new particle with the features of the elusive Higgs boson on 4th July 2012. We report evidence for non-trivial spatio-temporal patterns in user activities at individual and global level, such as tweeting, re-tweeting and replying to existing tweets. We provide a possible explanation for the observed time-varying dynamics of user activities during the spreading of this scientific "rumor". We model the information spreading in the corresponding network of individuals who posted a tweet related to the Higgs boson discovery. Finally, we show that we are able to reproduce the global behavior of about 500,000 individuals with remarkable accuracy.
Overview of the Dataset
The study analyzes nearly one million Higgs-related tweets and the follower network of their authors, revealing heterogeneous degree structure and degree correlations. Tweet activity increased sharply around the CERN announcement.
- Dataset: 985,590 tweets posted from 1–7 July 2012 formed a network of 456,631 nodes and 14,855,875 directed follower edges.The dataset contained tweets matching lhc, cern, boson, or higgs; 70,838 users were discarded because of inaccessible information.
- Network structure: For in-degree above kin ≈100, the network follows a power law with exponent −2.2, while lower in-degree values do not satisfy the scaling relation.
- Network structure: The network is degree-correlated and disassortative, with an assortative index of about −0.14 rather than approximately zero.Users with many links tend to connect preferentially to users with fewer connections.
Spatio-temporal Analysis
The Higgs announcement produced sharp, global changes in Twitter activity across time, space, and user-level behavior. Before and after the event, activity was bursty and geographically structured; during it, tweeting became frenetic and globally distributed, while distinct timing patterns appeared for tweets, replies, and re-tweets.
- Macroscopic activity: 36,000 tweets/hour marked the beginning of Period IV, rising from approximately 36 tweets/hour at the beginning of Period I before increasing by more than one order of magnitude after the announcement.The rate slowly decreased during the following days.
- Macroscopic activity: The announcement made Twitter activity truly global, whereas before and after it, the most active countries were primarily European and American.The paper associates this geographic concentration with the presence of scientists in those regions.
- Macroscopic activity: During the announcement, tweets were likely to follow one another within two seconds and no more than six seconds, indicating frenetic global activity.Before and after the event, inter-tweet-time distributions instead had long tails with many tweets within seconds and fewer within minutes.
- Macroscopic activity: During the announcement, tweets from any part of the world arrived within two seconds without a specific spatial pattern; afterward, activity became temporally similar to before the event but remained globally distributed.Before the event, consecutive tweets were more likely within 20 km and generally within eight seconds.
- Microscopic activity: Before and after the event, inter-tweet times followed P(τ) ∝ τ^-α with α ≈ 1 across three decades, from one minute to one day.This pattern is consistent with bursts of rapid activity separated by longer inactive periods.
- Microscopic activity: During the event, inter-tweet times followed a lognormal distribution with μ = 5.627±0.008 and σ = 1.742±0.006, described as fitting the observed activity with remarkable accuracy.The paper links this behavior to possible cascading triggered by tweets in users’ social neighborhoods.
- Microscopic activity: Reply inter-arrival times followed P(τ) ∝ τ^-1.2 from a few minutes to one day, while re-tweet times followed P(τ) ∝ τ^-0.8 with a cutoff scale τ0 ≈ 11 hours.The re-tweet distribution was modeled with a power law multiplied by an exponential cutoff.
Rumor Spreading
The paper models Higgs-boson rumor spreading on Twitter through user activation, network structure, and social cascading. Period-specific activation and de-activation models reproduce observed dynamics, including rapid growth and decline around the announcement.
- User states and network spreading: The analysis distinguishes active and non-active users and examines information spreading across the Twitter network through social cascading.Active users are those tweeting about the Higgs boson; refined dynamics allow users to become non-active again.
- Network structure: The k-core analysis reveals an inhomogeneous network with hubs distributed across shells, indicating non-trivial correlations and no apparent global hierarchy.The visualization uses vertex size for degree and color for k-coreness, with a 10% network sample.
- Activation without de-activation: The no-deactivation model treats active-user growth as monotonic and fits separate temporal ranges with a constant activation rate.The fitted activation rate rises from about one user per minute on 1 July to about 519 users per minute in the final period.
- Model assumptions: The model neglects neighborhood degree correlations as a simplifying assumption based on scale-free spreading theory.The assumption is applied because the network exhibits a scale-free degree distribution with exponent 2.5 for k > 200.
- Activation and de-activation dynamics: Observed active-user density rises rapidly around the conference, then decreases by about 40% in the following hour before temporarily stabilizing.During Period IV, five sub-periods show increasing activity followed by decline; the first decline has time scale τ ≈1.13 hours.
- Activation and de-activation dynamics: Coupled activation and de-activation equations reproduce the observed Period IV dynamics, with later decay time scales ranging from 3 to 17 hours.The model varies β, λ, and τ across five sub-periods and selects parameters minimizing χ2.
Discussion
The study tracks Twitter activity around the Higgs boson-like particle announcement, revealing distinct spatial-temporal dynamics and modeling the collective spread across hundreds of thousands of users.
- The study monitored Twitter user activity before, during, and after the announcement of a Higgs boson-like particle.
- Before the announcement, tweets were concentrated among nearby users and occurred within seconds; during the event, activity became faster and spatially diffuse.Afterward, activity was less frenetic, with users participating from across the geographic network.
- The authors modeled information spreading with a variable activation rate in a heterogeneous network and reproduced the collective behavior of about 500,000 users with remarkable accuracy.The model incorporates social reinforcement through the active or non-active states of users’ neighbors.
- The proposed framework may be applicable to other spreading processes on online and physical social networks.
Methods
The methods combine a keyword-filtered Twitter dataset, geographic processing, and a follower-network representation to study temporal information spreading.
- The authors report no missing tweets during collection and therefore consider the dataset complete for the stated search criteria.
- The dataset includes tweets retrieved with selected Higgs-related keywords and hashtags, excluding terms that produced a non-negligible amount of irrelevant content.
- Geographic names in Twitter profiles were converted into locations using the Google Geocoder API.
- The study uses the follower network rather than retweet or mentions networks to model the fraction of users involved over time.The choice reflects Twitter’s home timeline, where messages from social contacts can trigger activity regardless of reciprocal interactions.
- The social graph was constructed by retrieving each participating user’s following list, assuming collection was faster than significant changes in the full graph.