Computational Social Science


Human behavior at individual level is fascinating but can be barely understood or predicted. Instead, when a large collection of actors interact with each other, emergent collective behavior is observed, with predictable patterns.

A dynamical visualization of the multiplex interaction among users watching an X-Factor (Italy) competition. Users are nodes, color-coded by the community they belong to, while interactions are links, color-coded by the type (mention, reply, retweet). The result is a time-varying multiplex network.

Quantitative analysis of the infodemic phenomenon


During COVID-19, governments and the public are fighting not only a pandemic but also a co-evolving infodemic—the rapid and far-reaching spread of information of questionable quality. We analysed more than 100 million Twitter messages posted worldwide during the early stages of epidemic spread across countries (from 22 January to 10 March 2020) and classified the reliability of the news being circulated. We developed an Infodemic Risk Index to capture the magnitude of exposure to unreliable news across countries. We found that measurable waves of potentially unreliable information preceded the rise of COVID-19 infections, exposing entire countries to falsehoods that pose a serious threat to public health. As infections started to rise, reliable information quickly became more dominant, and Twitter content shifted towards more credible informational sources. Infodemic early-warning signals provide important cues for misinformation mitigation by means of adequate communication strategies.

Figure: The infodemic risk of each country, aggregated over time, is colour-coded on the map. The panels show the evolution of risk over time for a sample of countries; the bars indicate the partial contributions of verified and unverified users to the overall risk and the dashed lines represent the cumulative mean of the IRI at a given day d (computed as the ratio between the cumulative sum of the daily IRI in the days between 22 January and d, and the number of days between these two dates). Risk evolution for the whole world is also shown, demonstrating an overall decrease of risk over time (bottom middle panel, where the grey line represents a LOESS regression with R2 = 0.29). The markers horizontally aligned at the top of each panel indicate the daily confirmed epidemiological cases, with their number encoded by the markers’ sizes (Venezuela does not contain epidemiological markers as no confirmed cases were reported at the time of the anaysis). Map made with public domain Natural Earth data.


References:

   

Human-bot interaction and the spread of misinformation Learn more »


Analyzing large-scale social data collected during the Catalan referendum for independence on October 1 2017, consisting of nearly 4 millions Twitter posts generated by almost 1 million users, we identify the two polarized groups of Independentists and Constitutionalists and quantify the structural and emotional roles played by social bots. We show that bots act from peripheral areas of the social system to target influential humans of both groups, mostly bombarding Independentists with negative and violent contents, sustaining and inflating instability in this online society. These results quantify the potential dangerous influence of political bots during voting processes.




A more recent analysis of the same data set and of the online activity during the Yellow Vests protests, reveals that bots play an important role in attracting human attention, but their centrality and visibility are way smaller than verified (media) accounts in spreading information.




Centrality in the RT network. Panel A shows the network built with the Yellow Vests data; panel B shows the network built with the Catalan Referendum data. The boxplots summarize the values obtained from permutations of the data where the category labels were randomly reshuffled across accounts. The observed centrality of media accounts is significantly higher than expected by chance in both mobilizations. Human accounts, on the other hand, receive significantly fewer RTs. The axes preserve different scales to allow visual identification of distance between permutations and observed values

References:

   

Influence of human-bot interaction during voting events Learn more »


It has been shown that one of such mechanisms is based on fabricating armies of automated agents that are known as social bots. Using the recent Italian elections as an emblematic case study, here we provide evidence for the existence of a special class of highly influential users, that we name "augmented humans". They exploit bots for enhancing both their visibility and influence, generating deep information cascades to the same extent of news media and other broadcasters. Augmented humans uniformly infiltrate across the full range of identified clusters of accounts, the latter reflecting political parties and their electoral ranks.


Figure: Information cascades during Italian elections. a) Heatmap of the number of users initiating information cascades, as a function of the size of their social neighborhood (Followers) and the size of the generated cascade; b) Scatter plot of the same data, with points encoding users. Color encodes bot/human classification and size encodes cascade’s diameter; c) As in a) but considering cascade rate, defined by the ration between cascade size and its duration, vs. neighborhood size (left panels) and cascade size (right panels), for humans (top panels) and bots (bottom panels). The heatmap of cascade rate vs. neighborhood size allows one to identify 4 categories: hidden influentials, influentials, common users and broadcasters (see the text for further detail). Dashed lines indicate medians of structural and dynamical features in humans. Only cascades with at least 10 adopters are considered and, for heatmaps, the logarithm of the corresponding variables is considered.

References:

   

Unraveling the Origin of Social Bursts in Collective Attention


In the era of social media, every day billions of individuals produce content in socio-technical systems resulting in a deluge of information. However, human attention is a limited resource and it is increasingly challenging to consume the most suitable content for one’s interests. In fact, the complex interplay between individual and social activities in social systems overwhelmed by information results in bursty activity of collective attention which are still poorly understood. Here, we tackle this challenge by analyzing the online activity of millions of users in a popular microblogging platform during exceptional events, from NBA Finals to the elections of Pope Francis and the discovery of gravitational waves. We observe extreme fluctuations in collective attention that we are able to characterize and explain by considering the co-occurrence of two fundamental factors: the heterogeneity of social interactions and the preferential attention towards influential users. Our findings demonstrate how combining simple mechanisms provides a route towards understanding complex social phenomena.

Figure: Social bursts of collective attention during exceptional events. (A) Volume of activity in tweets/minute (y-axis) as a function of time (x-axis, measured in hours) observed in the microblogging platform Twitter and measured during special events (Pope Francis’ election in 2013, the discovery of gravitational waves in 2016, the Cannes Film Festival in 2013, and the 50th anniversary of Martin Luther King’s most famous speech in 2013). (B) Bursts decay either instantaneously (top) or with some characteristic relaxation dynamics (bottom)


Figure: Fluctuation analysis of social bursts during collective attention. Spikiness S – Eq. (3)– is plotted against number of tweets – NT,w – for two social activities (replies and retweets) during four exceptional events. Each dot is the result (empirical data) obtained in a time window w of size 20 minutes (i.e., NT,w = 1000 indicates an average of 50 posts per minute). Shaded areas indicate the 90% confidence around the expected S obtained simulating our model in the three scenarios: (i) homogenous social structure with uniformly distributed attention (“Hom.”); (ii) social structure obtained from preferential attachment with uniformly distributed attention (“Het.”); (iii) social structure obtained from preferential attachment with preferential attention (“Het. and Atten.”).


References:

   

What is the difference between moderated and unmoderated social systems?


Online social networks are the perfect test bed to better understand large-scale human behavior in interacting contexts. Although they are broadly used and studied, little is known about how their terms of service and posting rules affect the way users interact and information spreads. Acknowledging the relation between network connectivity and functionality, we compare the robustness of two different online social platforms, Twitter and Gab, with respect to banning, or dismantling, strategies based on the recursive censor of users characterized by social prominence (degree) or intensity of inflammatory content (sentiment). We find that the moderated (Twitter) vs. unmoderated (Gab) character of the network is not a discriminating factor for intervention effectiveness. We find, however, that more complex strategies based upon the combination of topological and content features may be effective for network dismantling. Our results provide useful indications to design better strategies for countervailing the production and dissemination of anti-social content in online social platforms.


Figure: (A) Sketch showing the mapping between the online dynamics and the reconstructed behavioral networks. Attacks are identified with the removal of agents and might split the network in smaller components. In our case, two sub-networks appear once the actor B is attacked: {𝐴,𝐷,𝐸} and {𝐶}. The overall robustness of a system is related to its capacity of maintaining a component as large as possible when suffering attacks. (B) and (C): Curves of the largest connected component under degree-based attacks in the network of replies and the network of mentions, respectively. “Moderated” corresponds to Twitter and “Unmoderated” to Gab. The above bubble plots show the connected components remaining at the percolation point. The size of the bubbles is related to the logarithm of the component size.


Figure: (A) Sketches showing the k-core structure of a toy network and heat maps with a decentralized and centralized network organization. The patterns exemplified in the heat maps A1 to A4 represent scenarios in which the connectivity is involving predominantly a single k-shell. In A1 most connections are between nodes of the central core (𝐾=4) and other central shells (as in assortative networks). In A2 most connections are between the central core and nodes of external shells (as in dis-assortative networks). In A3 and A4 most connections involve the most external shell (𝐾=1). This can also happen in an assortative way (A3), where most communication will be between isolated couples of nodes of the external shells, or a dis-assortative way (A4) where most communication is between a large external shell of leaves and the central shells (core-periphery structure). The patterns of heat maps A5 and A6 describe instead networks where connections are distributed between multiple shells in such a way that nodes in a k-shell are mostly connected to the same shell or to shells below and above in the hierarchy. They can be obtained by overlapping single shell effects similar to that of A1. In A5, connections are homogeneously distributed across shells, while in A6 connections are more abundant in the central shells. In (B) and (C), k-shell decomposition and associated heat map for Twitter and Gab, respectively. The distribution of nodes and their degrees across the different k-shells can be appreciated by representing the different nodes in concentric circumferences (plotted above using LaNet-vi52). In the heat maps we focus instead on the connectivity. In the light of the pattern exemplified above, we observe how the Twitter reply and mention and the Gab reply networks can be seen as an overlap of A1 and A2 (with the A3 patterns also playing a minor role in Twitter). The Gab mention network differs notably, and can be seen as an overlap of the A2, A4 and A6 patterns: more communication is thus seen in the intermediate shells (A6) and between marginal and central nodes (A4).


References:

   

Identifying influent actors in multiplex systems


A challenging problem is to identify them in networked systems characterized by different types of interactions, forming interconnected multilayer networks. Here we describe a mathematical framework that allows us to calculate centrality in such networks and rank nodes accordingly, finding the ones that play the most central roles in the cohesion of the whole structure, bridging together different types of relations. These nodes are the most versatile in the multilayer network. We investigate empirical interconnected multilayer networks and show that the approaches based on aggregating—or neglecting—the multilayer structure lead to a wrong identification of the most versatile nodes, overestimating the importance of more marginal agents and demonstrating the power of versatility in predicting their role in diffusive and congestion processes.


Figure: The concept of versatility is fundamental to remove degeneracy due to aggregated information. It is used to identify actors who play a key role during multilayer network dynamics.

References:

   

Robustness os social-ecological systems to climate and social changes read more


Social capital ties are ubiquitous in modern life. For societies with people and landscapes tightly connected, in variable or marginal ecosystems, and with unreliable market sectors, social relations are critical. Each relation is a potential source of food, information, cash, labor, or expertise. Here, we present an analysis of multiplex, directed, and weighted networks representing actual flows of subsistence-related goods and services among households in three remote indigenous Alaska communities exposed to both extreme climate change and industrial development. We find that the principal challenge to the robustness of such communities is the loss of key households and the erosion of cultural ties linked to sharing and cooperative social relations rather than resource depletion.

#
Figure: Robustness and vulnerability of multilayer social-ecological networks to plausible scenarios of change. Top left: study areas in northern Alaska. Bottom Left: Robustness of the multilayer network of the three communities (band) to random and targeted perturbations. Targeted perturbation hit the most important HH or layers, while random perturbation are averaged over 1000 different random removal of HH or layers. Bottom right: Vulnerability of the three community (on average) to the targeted and random removal of layers and nodes. Nodes are removed in case of the HHL (house hold loss scenario), while layers are removed in case of the SS (social shift), RD (resource depletion), TRD (terrestrial resource depletion) and RRD (riverine resource depletion). Social Shift and Household Loss are the most impactful perturbations independent from the community under study.


References:

   

Information diffusion in socio-technical systems Learn more »


We provide a possible explanation for the observed time-varying dynamics of user activities during the spreading of news about the Higgs boson discovery. We model the information spreading in the corresponding network of individuals who posted a related tweet and show that we are able to reproduce the global behavior of about 500,000 individuals with remarkable accuracy.

#
Figure: Number of tweets per second as a function of time during the period of data collection.
The curves correspond to tweets containing only the CERN, Higgs, LHC keywords and at least one of them, respectively.


References: