Social media are interactive technologies that facilitate the creation and sharing of information, ideas, interests, and other forms of expression through virtual communities and networks. While challenges to the definition of social media arise due to the variety of stand-alone and built-in social media services currently available, there are some common features:

Social media are interactive Web 2.0 Internet-based applications.

User-generated content—such as text posts or comments, digital photos or videos, and data generated through all online interactions—is the lifeblood of social media.

Users create service-specific profiles for the website or app that are designed and maintained by the social media organization.

Social media helps the development of online social networks by connecting a

The Role of Compromised Accounts in Social Media Manipulation

Tugrulcan Elmas

In recent years we have seen a marked increase in disinformation including as part of a strategy of so-called hybrid warfare. Adversaries not only directly spread misleading content but manipulate social media by employing sophisticated techniques that exploit platform vulnerabilities and avoid detection. It is getting increasingly important to analyze social media manipulation to better understand, detect and defend public dialogue against it.In this thesis, we contribute to the research on social media manipulation by describing and analyzing how adversaries employ compromised social media accounts. We begin by providing a background of social media: we describe the mechanisms and the influence of the platforms to better understand why the adversaries target them. We then give a detailed overview of social media manipulation, and the techniques to detect and counter it. Next, we discuss our contributions in this thesis: 1) an extensive analysis of an attack on social media algorithms using compromised accounts, 2) a study of the implications of compromised bots for bot research through the characterization of retweet bots, 3) a detection method to find compromised accounts that are later repurposed.Firstly, we uncover and analyze a previously unknown, ongoing astroturfing attack on the popularity mechanisms of social media platforms: ephemeral astroturfing attacks. In this attack, a chosen keyword or topic is artificially promoted by coordinated and inauthentic activity to appear popular. Crucially, this activity is removed as part of the attack which facilitates using compromised accounts that are still managed by their original owners. We detected over 19,000 unique fake trends promoted by over 108,000 accounts. Trends astroturfed by these attacks account for at least 20% of the top 10 global trends. We created a Twitter bot to detect the attacks in real-time and inform the public. Secondly, we study the implications of compromised accounts to bot research. We do this by characterizing retweet bots that have been uncovered by purchasing retweets from the black market. We determine that those accounts were compromised as they observe anomalous behavior, or self-state that they are hacked. We then analyze their differences from human-controlled accounts. From our findings on the nature and life-cycle of retweet bots, we point out several inconsistencies between the retweet bots used in this work and bots studied in prior works. Our findings challenge some of the fundamental assumptions related to bots and in particular how to detect them.Thirdly, we define, describe, and provide a detection method for misleading repurposing, in which an adversary changes the identity of a potentially compromised social media accounts via, among other things, changes to the profile attributes in order to use them for a new purpose while retaining their followers. We propose a methodology to flag repurposed accounts, and detected over 100,000 such accounts. We also characterize repurposed accounts and found that they are more likely to be repurposed after a period of inactivity and deleting old tweets. We present a tool to root out accounts that became popular and repurposed later.Our work is significant in presenting how breaches of user security jeopardize platform security and public dialogue. Furthermore, it enhances the knowledge of how the bots and troll accounts work and aid platforms and researchers in building new solutions.

EPFL2022

Understanding Eating And Drinking In Context From Crowdsourced Data

Thanh Trung Phan

With the increasing rate of urbanization, understanding food and beverage consumption, including alcohol drinking behaviour with its consequences, is relevant in such a megacity in the future. Especially, investigation of alcohol drinking is necessary for uncovering more drinking activities of young people in context, and providing more scientific references to authorities to improve public policies, in particular in the domain of public health. In previous research, data were collected a posteriori in face-to-face interviews or by using questionnaires. Thanks to the development of feature phones, an evolution of more traditional methods in ubiquitous food and alcohol research could have collected survey data via SMS. However, these methods have limitations, including a low recall and an expensive scaling up. On the other hand, the adoption of smartphones and social media is opening new channels for investigating behaviours by collecting fine-grained in-situ data, following methodologies from social sciences, and using advanced technologies from computer sciences. Recently, crowdsourcing is a new paradigm that consists in using the inputs from a great number of people to facilitate and accelerate large scale data collection from broad samples, compared with traditional methods. In addition, mobile crowdsourcing, a form of crowdsourcing, has enormous potential in collecting in-situ data by taking advantage of embedded sensors, cameras, and being equipped with Internet connection. In this dissertation, we investigate the drinking and eating behaviour of young people in Switzerland, based on crowdsourcing data including records and metadata from mobile sensors (mobile crowdsensing) and data shared on social networks. Our contributions are three-fold. First, we conduct data analyses that uncover generic food and drink consumption on Instagram and reveal two types of drinking practices (casual and heavy drinking) on social media. This analysis provides an initial snapshot of the food consumption and alcohol drinking practices based on the way they appear online, and creates a preliminary alcohol consumption model that is developed in the rest of the dissertation. Second, we use mobile crowdsensing data, annotated a posteriori, both to identify heavy drinking and to understand the characteristics of private spaces (including ambiances) and activities (including drinking activities) of young people in the weekend nights in Switzerland. These results show how mobile crowdsensing data can be used to better understand and predict alcohol drinking practices and ambiances in private spaces. Third, we combine mobile crowdsensing data with social media to retrieve the multi-factorial characteristics of drinking events depending on the type of beverage (multiple alcoholic and non-alcoholic categories) based on images features and contextual cues from individual and joint data sources. This result indicates the feasibility of using, individually or combined, data from mobile crowdsensing and social networks in discriminating drinking behaviour. This is a promising sign towards the development of a system that uses machine learning for self-monitoring of alcohol consumption. This dissertation, by combining the advanced machine learning of computer science and literature of social science, demonstrates the relevance of using a multidisciplinary approach to investigate social behaviours in urban areas.

EPFL2020

On the use of applied machine learning and digital infrastructure to leverage social media data in health and epidemiology

Martin Mathias Müller

The quantification of population-level health behaviors is crucial for guiding public health policy. However, traditional methods for measuring such health behaviors have several short- comings. In recent years social media data has been successfully used to measure health behaviors and may be used as a low-cost and real-time addition to traditional data sources. Methods from the field of natural language processing are increasingly used to automatically process, filter and categorize the rapidly growing amount of publicly available social media data. However, a number of methodological challenges limit the rate at which we can generate insight from such data. In this work I will argue that long-term investment into digital infrastructure and open source tooling is required in order to overcome these challenges. In chapter 2 we introduce the Crowd- breaks platform which is the basis of this thesis. Crowdbreaks is an open source framework for real-time data collection, continuous crowdsourced annotation, and continuous re-training of machine learning classifiers. In contrast to traditional research workflows, projects on Crowdbreaks run over an extended period of time, allowing for the observation of health trends over multiple years while keeping algorithms up-to-date. In chapter 3 we quantify the occurrence of concept drift in vaccine-related Twitter data, which further validates the need for the Crowdbreaks platform. In chapter 4 we use the Crowdbreaks platform to trace sentiment towards the novel gene-editing technology CRISPR/Cas9 back to its first application in 2013 and investigate how public opinion may have been affected in context of recent scandals sur- rounding the technology. In chapter 5 we turn our attention to the COVID-19 pandemic and analyze who was speaking and who was heard in the early months of the pandemic. Chapter 6 builds on this work and explores the dynamics of Twitter communities during the COVID-19 pandemic. Lastly, in chapter 7 we introduce COVID-Twitter-BERT, a domain-specific language model which has been used in various downstream natural language processing applications on COVID-19-related Twitter data.

EPFL2021