Online Social Networks and Media最新文献

With proliferation of user generated contents in social media platforms, establishing mechanisms to automatically identify toxic and abusive content becomes a prime concern for regulators, researchers, and society. Keeping the balance between freedom of speech and respecting each other dignity is a major concern of social media platform regulators. Although, automatic detection of offensive content using deep learning approaches seems to provide encouraging results, training deep learning-based models requires large amounts of high-quality labeled data, which is often missing. In this regard, we present in this paper a new deep learning-based method that fuses a Back Translation method, and a Paraphrasing technique for data augmentation. Our pipeline investigates different word-embedding-based architectures for classification of hate speech. The back translation technique relies on an encoder–decoder architecture pre-trained on a large corpus and mostly used for machine translation. In addition, paraphrasing exploits the transformer model and the mixture of experts to generate diverse paraphrases. Finally, LSTM, and CNN are compared to seek enhanced classification results. We evaluate our proposal on five publicly available datasets; namely, AskFm corpus, Formspring dataset, Warner and Waseem dataset, Olid, and Wikipedia toxic comments dataset. The performance of the proposal together with comparison to some related state-of-art results demonstrate the effectiveness and soundness of our proposal.

Studies on online hate speech have mostly focused on the automated detection of harmful messages. Little attention has been devoted so far to the development of effective strategies to fight hate speech, in particular through the creation of counter-messages. While existing manual scrutiny and intervention strategies are time-consuming and not scalable, advances in natural language processing have the potential to provide a systematic approach to hatred management. In this paper, we introduce a novel ICT platform that NGO operators can use to monitor and analyse social media data, along with a counter-narrative suggestion tool. Our platform aims at increasing the efficiency and effectiveness of operators’ activities against islamophobia. We test the platform with more than one hundred NGO operators in three countries through qualitative and quantitative evaluation. Results show that NGOs favour the platform solution with the suggestion tool, and that the time required to produce counter-narratives significantly decreases.

In the last few years, thanks to the emergence of Web 2.0, social media has made the concept of online live events possible. Users participate more and more in long-running recurring events in social media by sharing their experiences and desires. In the last few years, thanks to the emergence of Web 2.0, social media has made the concept of online live events possible. Users participate more and more in long-running recurring events in social media by sharing their experiences and desires. This work introduces long-running live events (LRLEs), as a type of activity that span physical spaces and digital ecosystems, including social media. LRLEs encompass several individuals, organizations, and brands collaborating/competing in the same event. This provides unprecedented opportunities to understand the dynamics and behavior of event-oriented participation, through collection and analysis of data of user behaviors enabled by the Web platform, where most of the digital traces are left by users. What makes this setting interesting is that the behaviors that are traced are not focused only on one individual brand or organization, and thus allows one to understand and compare the respective roles and influence in a defined setting. In this paper we provide a high-level and multi-perspective roadmap to mine, model, and study LRLEs. Among the various aspects, we develop a multi-modal approach to solve the problem of post popularity prediction that exploits potentially influential factors within LRLE. We employ two methods for implementing feature selection, together with an automated grid search for optimizing hyper-parameters in various regression methods.

Cyberaggression has been studied in various contexts and online social platforms, and modeled on different data using state-of-the-art machine and deep learning algorithms to enable automatic detection and blocking of this behavior. Users can be influenced to act aggressively or even bully others because of elevated toxicity and aggression in their own (online) social circle. In effect, this behavior can propagate from one user and neighborhood to another, and therefore, spread in the network. Interestingly, to our knowledge, no work has modeled the network dynamics of aggressive behavior. In this paper, we take a first step towards this direction by studying propagation of aggression on social media using opinion dynamics. We propose ways to model how aggression may propagate from one user to another, depending on how each user is connected to other aggressive or regular users. Through extensive simulations on Twitter data, we study how aggressive behavior could propagate in the network. We validate our models with crawled and annotated ground truth data, reaching up to 80% $AUC$, and discuss the results and implications of our work.

Online Social Networks (OSN) have become a central means of communication and interaction between people around the world. The essence of privacy has been challenged through the past two decades as technological advances enabled benefits and social visibility to active members that share content in online communities. While OSN users share personal content with friends and colleagues, they are not always fully aware of the potential unintentional exposure of their information to various people including adversaries, social bots, fake users, spammers, or data-harvesters. Preventing this information leakage is a key objective of many security models developed for OSNs including Access Control, Relationship based models, Trust based models and Information Flow control. Following previous research, we assert that a combined approach is required to overcome the shortcoming of each model. In this paper we present a new model to protect users' privacy that is composed of three main phases addressing three of its major aspects: trust, role-based access control and information flow. This model considers a user's sub-network and classifies the user's direct connections to roles. It relies on public information such as total number of friends, age of user account, and friendship duration to characterize the quality of the network connections. It also evaluates trust between a user and members of the user's network to estimates if these members are acquaintances or adversaries based on the paths of the information flow between them. Finally, it provides more precise and viable information sharing decisions and enables better privacy control in the social network. We have evaluated our model with extensive experiments using both synthetic and real users' networks to demonstrate its ability to provide a naïve user with a good means of privacy protection. We have validated separately every phase of the model and examined the decisions obtained by two different approaches. The results show a strong correlation between the decisions made by the algorithm and the users' decisions.

Emojis or ‘picture characters’ have become ubiquitous in modern-day digital communication, including social media sharing and smartphone texting. Despite this ubiquity, many questions remain about their usage, especially with respect to global variations in language and country. These questions are important, in part because they reveal how people communicate digitally on social platforms, but also because they provide a lens through which different regions and cultures can be studied. In this paper, we conduct a principled, quantitative study to understand emoji usage in terms of linguistic and country correlates. Our study involves 30 languages and countries each, and is conducted over tens of millions of tweets collected from the Twitter decahose over an entire month. Drawing on both statistical measures and information theory, our results reveal that, not only does emoji usage have strong dependencies at both the language and country level, but that some languages and countries are much more constrained in the diversity of their emoji usage. However, we also discover that the ‘popularity’ of emojis, both globally and within the context of a given language, follows a robust and invariant trend that emerges fairly quickly (over just a day’s worth of data) and cannot be explained either by a power-law or Heap’s law-like distribution.

Social media networks have revolutionized social science research. Yet, a lack of comparative empirical analysis of these networks leave social scientists with little knowledge on the role that contextual factors play in the formation of social relations. In this paper we perform a large-scale comparison of parliamentary Twitter networks in 12 countries to improve our understanding of the influence of the country’s democratic system on network behavior and elite polarization. One year of Twitter data was collected from all members of the parliament and government in these countries, which resulted in around two million tweets by almost 6000 politicians. Social network analysis of the Twitter interactions indicates that consensual democracies are characterized by more dense parliamentary relations but also higher hierarchy and fragmentation compared to majoritarian systems. Secondly, parliaments with a high effective number of parties are more cooperative, which results in higher inter-party relations. Next to that, we show differences in the followers, mentions, and retweets networks that hold across all countries and political systems. Our empirical results correspond to established theoretical insights and highlight the relevance of institutional context as well as the platform characteristics when conducting social media research. With this research we demonstrate the importance and the opportunities of social network analysis for comparative research.

The adversarial impact of the Covid-19 pandemic has created a health crisis globally all over the world. This unprecedented crisis forced people to lockdown and changed almost every aspect of the regular activities of the people. Thus, the pandemic is also impacting everyone physically, mentally, and economically, and it, therefore, is paramount to analyze and understand emotional responses during the crisis affecting mental health. Negative emotional responses at fine-grained labels like anger and fear during the crisis might also lead to irreversible socio-economic damages. In this work, we develop a neural network model and train it using manually labeled data to detect various emotions at fine-grained labels in the Covid-19 tweets automatically. We present a manually labeled tweets dataset on COVID-19 emotional responses along with regular tweets data. We created a custom Q&A roBERTa model to extract phrases from the tweets that are primarily responsible for the corresponding emotions. None of the existing datasets and work currently provide the selected words or phrases denoting the reason for the corresponding emotions. Our classification model outperforms other systems and achieves a Jaccard score of 0.6475 with an accuracy of 0.8951. The custom RoBERTa Q&A model outperforms other models by achieving a Jaccard score of 0.7865. Further, we present a historical emotion analysis using COVID-19 tweets over the USA including each state level analysis.

Social media play an important role in the daily life of people around the globe and users have emerged as an active part of news distribution as well as production. The threatening pandemic of COVID-19 has been the lead subject in online discussions and posts, resulting to large amounts of related social media data, which can be utilised to reinforce the crisis management in several ways. Towards this direction, we propose a novel framework to collect, analyse, and visualise Twitter posts, which has been tailored to specifically monitor the virus spread in severely affected Italy. We present and evaluate a deep learning localisation technique that geotags posts based on the locations mentioned in their text, a face detection algorithm to estimate the number of people appearing in posted images, and a community detection approach to identify communities of Twitter users. Moreover, we propose further analysis of the collected posts to predict their reliability and to detect trending topics and events. Finally, we demonstrate an online platform that comprises an interactive map to display and filter analysed posts, utilising the outcome of the localisation technique, and a visual analytics dashboard that visualises the results of the topic, community, and event detection methodologies.

Thanks to platforms such as Twitter and Facebook, people can know facts and events that otherwise would have been silenced. However, social media significantly contribute also to fast spreading biased and false news while targeting specific segments of the population. We have seen how false information can be spread using automated accounts, known as bots. Using Twitter as a benchmark, we investigate behavioural attitudes of so called ‘credulous’ users, i.e., genuine accounts following many bots. Leveraging our previous work, where supervised learning is successfully applied to single out credulous users, we improve the classification task with a detailed features’ analysis and provide evidence that simple and lightweight features are crucial to detect such users. Furthermore, we study the differences in the way credulous and not credulous users interact with bots and discover that credulous users tend to amplify more the content posted by bots and argue that their detection can be instrumental to get useful information on possible dissemination of spam content, propaganda, and, in general, little or no reliable information.