Online Social Networks and Media最新文献

Social media platforms are increasingly used to convey advertising campaigns for products or services. A key issue is to identify an appropriate set of influencers within a social network, investing resources to get them to adopt a product. Influence maximization is an optimization problem that aims at finding a small set of users that maximize the spread of influence in a social network. In this paper we propose an influence maximization algorithm, named Weighted Artificial Bee Colony (WABC), that is based on a bio-inspired technique for identifying a subset of users which maximizes the spread. The proposed algorithm has been applied to a case study that analyzes the propagation of information among Twitter users during the Constitutional Referendum held in Italy in 2016. Our analysis is aimed at identifying the main influencers of the $yes$ and $no$ factions, and deriving the main information diffusion strategies of each faction during the political campaign. WABC outperformed ranking-proxy techniques based on classical centrality measures, i.e., PageRank, Rank and Degree. Even compared to DIRIE, which exploits a more complex algorithm, WABC was able to find a more accurate set of users which allows to maximize the spread in almost all the considered configurations.

During the coronavirus disease 2019 (COVID-19) pandemic, the video-sharing platform YouTube has been serving as an essential instrument to widely distribute news related to the global public health crisis and to allow users to discuss the news with each other in the comment sections. Along with these enhanced opportunities of technology-based communication, there is an overabundance of information and, in many cases, misinformation about current events. In times of a pandemic, the spread of misinformation can have direct detrimental effects, potentially influencing citizens' behavioral decisions (e.g., to not socially distance) and putting collective health at risk. Misinformation could be especially harmful if it is distributed in isolated news cocoons that homogeneously provide misinformation in the absence of corrections or mere accurate information. The present study analyzes data gathered at the beginning of the pandemic (January–March 2020) and focuses on the network structure of YouTube videos and their comments to understand the level of informational homogeneity associated with misinformation on COVID-19 and its evolution over time. This study combined machine learning and network analytic approaches. Results indicate that nodes (either individual users or channels) that spread misinformation were usually integrated in heterogeneous discussion networks, predominantly involving content other than misinformation. This pattern remained stable over time. Findings are discussed in light of the COVID-19 “infodemic” and the fragmentation of information networks.

During the 2016 US presidential election, we witnessed a polarized population and an election outcome that defied the predictions of many media sources. In this study, we conducted a follow-up on political view migration through tracking Twitter users’ account activity. The study was conducted by following a set of Twitter users over a four year period. Each year, Twitter user activities were collected and analyzed by our novel cross-account data mining algorithm. This algorithm through multiple iterations computes a numerical political score for each user based on their connection to other users and hashtags. We identified a set of seed users and hashtags using prominent political figures and movements to bootstrap the algorithm. The political score distribution demonstrates a divided population on political views. We also observed that users are more moderate in years close to elections (2017 and 2020) compared to years of none election (2018 and 2019). There is an overall migration trend from conservatives to progressives during the four years. This change in scores across the four year time frame suggests a unique political cycle exclusive to Donald Trump’s unprecedented presidential term. Our results in a broad sense portray the potential capabilities of a data collection and scoring algorithm that detected a noticeable political migration and describes the broad social characteristics of certain politically aligned users on social media platforms.

We studied international migrations of researchers, scientists, and academics, to better understand the so-called “brain drain” phenomenon, if and how it can be measured, and how it changes over time. We discuss why some trivial measures can be ineffective, and as a consequence, we built the global scientific migration network to identify the most important countries involved in the mobility of scholars, and to study their role at a local and a global scale.

For such a purpose, we analysed a temporal directed weighted network representing scientists moving from one country to another, from 2000 to 2016, built on top of 2.8 million ORCID public profiles. With the support of the well-known HITS algorithm, we found hubs and authorities to study the interplay between providing and attracting researchers from a global perspective, and its relationship to other structural features.

Our findings highlight the presence of a set of countries acting both as hubs and authorities, occupying a privileged position in the Scientific Migration Network, that is network of the scientific migrations, and having similar local characteristics, i.e., several neighbours with highly differentiated flows of researchers moving from/to them. However, it is striking that some of these countries have a predominant role over the others, and that we can easily observe countries that are extremely more attractive than others, as well as other countries that perform better as exporters than importers of scientists. It is also interesting that hubs and authorities scores can change over time, alongside with their relative discrepancy, and other network measures, suggesting that local and/or global policies can buck the trend.

Online firestorms pose a serious threat to companies and cause spontaneous information asymmetry between companies and social media users, which is part of the principal-agent theory. Corporate crisis management has already developed strategies to deal with firestorms, but these strategies are more effective if the company identifies a firestorm at an early stage. Therefore, we first identify literature-based characteristics of firestorms and quantify these using data-driven features in a multiple-case study approach based on Twitter data. Secondly, we identify per case the beginning of the firestorm and days with the least fluctuation in the number of posts as reference days. Finally, we compare the features between the starting points and the reference days to determine which features are significantly different. We could identify 24 features that change significantly at the beginning of a firestorm. This enables us to determine which features a company must pay particular attention to in order to detect a firestorm at an early stage. Likewise, we discuss these features in the context of the principal-agent theory with the use of social synchrony and crowd psychology to show how these features change information diffusion and contribute to information asymmetry.

Social networks such as Facebook, Twitter, Instagram play an important role in information diffusion. To understand how information is diffused in these social networks, it is important to examine users’ online activities and behaviors. In this work, we focus on Twitter and study the impact of users’ behaviors on their retweet activities (the major way of information diffusion on Twitter). We consider the topic preference, emotion and personality of a user as part of the user profile to represent their online behavior. The user profile can be built based on all their past tweets, retweets, or both. We propose two types of retweet prediction models, one is using classification algorithms, and the other is using matrix factorization algorithms. In the matrix factorization approach, we include behavior features into the basic factorization model through newly defined regularization terms. The experimental results show that in terms of the F1-score, our classification models based on user behavior related features provided 5%-9% improvement over baseline models and the matrix factorization model showed 4%-6% improvement over the baseline. We also find that by only considering the retweets, the data processing time is reduced while the prediction accuracy is comparable to the case when both retweets and tweets are included.

Influence propagation in social networks is a subject of growing interest. A relevant issue in those networks involves the identification of key influencers. These players have an important role on viral marketing strategies and message propagation, including political propaganda and fake news. In effect, an important way to fight malicious usage on social networks is to understand their properties, their structure and the way messages propagate.

This paper proposes a new index for analyzing message propagation in social networks, based on the network topological nature and the influential power of the message. The new index characterizes the strength of each node as a launcher of the message, dividing the nodes into launchers and non-launchers. This division is most evident when the viral power of the message is high. Together with other known metrics, launcher individuals can assist to select efficient influencers in a social network. For instance, instead of choosing a strong member according to its degree in the network (number of followers), we may previously select those belonging to the launchers group and then look for the lowest degree members contained therein. These members are probably cheaper (on financial incentives) but still guarantying almost the same influence effectiveness as the largest degree members.

We discuss this index using a number of real-world social networks available in known datasets repositories.

The rapid proliferation of misinformation and disinformation on the Internet has brought dire consequences upon societies around the world, fostering extremism, undermining social cohesion and threatening the democratic process. This impact can be attested by recent events like the COVID-19 pandemic and the 2020 US presidential election. The impact of misinformation has been so deep and wide that several authors characterize the present historic period as the “post-truth” era. Many recent efforts seek to contain the proliferation of misinformation by automating the identification of fake news through various techniques that exploit signals derived from linguistic processing of online content, analysis of message diffusion patterns, reputation lists, etc. In this paper we describe the design, implementation of, and experimentation with Check-It, a lightweight, privacy preserving browser plugin that detects fake-news. Check-It combines knowledge extracted from a variety of signals, and outperforms state-of-the-art methods on commonly-used datasets, achieving more than 90% accuracy, as well as a smooth user experience.

Instagram has been increasingly used as a source of information especially among the youth. As a result, political figures now leverage the platform to spread opinions and political agenda. We here analyze online discussions on Instagram, notably in political topics, from a network perspective. Specifically, we investigate the emergence of communities of co-commenters, that is, groups of users who often interact by commenting on the same posts and may be driving the ongoing online discussions. In particular, we are interested in salient co-interactions, i.e., interactions of co-commenters that occur more often than expected by chance and under independent behavior. Unlike casual and accidental co-interactions which normally happen in large volumes, salient co-interactions are key elements driving the online discussions and, ultimately, the information dissemination. We base our study on the analysis of 10 weeks of data centered around major elections in Brazil and Italy, following both politicians and other celebrities. We extract and characterize the communities of co-commenters in terms of topological structure, properties of the discussions carried out by community members, and how some community properties, notably community membership and topics, evolve over time. We show that communities discussing political topics tend to be more engaged in the debate by writing longer comments, using more emojis, hashtags and negative words than in other subjects. Also, communities built around political discussions tend to be more dynamic, although top commenters remain active and preserve community membership over time. Moreover, we observe a great diversity in discussed topics over time: whereas some topics attract attention only momentarily, others, centered around more fundamental political discussions, remain consistently active over time.

In recent years, the problem of unreliable content in social networks has become a major threat, with a proven real-world impact in events like elections and pandemics, undermining democracy and trust in science, respectively. Research in this domain has focused not only on the content but also on the accounts that propagate it, with the bot detection task having been thoroughly studied. However, not all bot accounts work as unreliable content spreaders (p.e. bot for news aggregation), and not all human accounts are necessarily reliable. In this study, we try to distinguish unreliable from reliable accounts, independently of how they are operated. In addition, we work towards providing a methodology capable of coping with real-world situations by introducing the content available (restricting it by volume- and time-based batches) as a parameter of the methodology. Experiments conducted on a validation set with a different number of tweets per account provide evidence that our proposed solution produces an increase of up to 20% in performance when compared with traditional (individual) models and with cross-batch models (which perform better with different batches of tweets).