Short Courses

Polarization in Social Networks: Concepts, Applications and Challenges

Bruno Hott (UFMG), Bruno Pereira Santos (UFMG), Túlio Loures (UFMG), Fabricio Benevenuto (UFMG), Pedro Vaz de Melo (UFMG)

The polarization measured in social networks has reflected the predisposition of society to the clash of ideas and the recent encouragement of political rivalry in the world. In this context, several questions are raised such as: Are people becoming more polarized? If so, what is the positive and negative impact of social media on this process? How to measure/predict polarizing potential of a post, comment or text? Among others. In the direction of the possible answers to these questions, several technical challenges will eventually need to be overcome, for example, how to train an algorithm to detect the polarization of a comment or text and how to implement tools to mitigate polarization on different social platforms. In this mini-course, the objective is to discuss the current scenario of research on polarization, through a critical overview of the area, its challenges and opportunities. To achieve these goals, the main concepts and definitions of polarization will be presented. The flow of data collection on polarization, its processing, analysis and knowledge extraction will also be presented. For the latter, a special focus will be given to a taxonomy proposal for polarization metrics in social networks. At the end, the challenges and opportunities found in the area of polarization in social networks will be discussed, as well as their impacts on society and the Internet as we know it.

Natural Language Processing in Social Media Texts: Fundamentals, Tools and Applications

Frances Santos (UNICAMP), Leandro Villas (UNICAMP), Thiago Henrique Silva (UTFPR), Jordan Kobellarz (UTFPR), Fábio Rezende de Souza (USP)

This mini-course has as main objective to present fundamentals and technologies in the area of Natural Language Processing (PLN – or NLP, from the English acronym of Natural Language Processing) for the development of applications through the exploration of social media texts written in English. . In this way, this mini-course prepares participants to: (1) know the main characteristics of texts shared by users on social media and how to collect them; (2) understand various NLP techniques to build and run NLP pipelines, including the steps of pre-processing, representation, modeling, knowledge extraction, semantic and emotional understanding, from social media texts; and (3) present possible applications that can benefit from the knowledge extracted from such texts.

Identification of Echo Chambers in Social Networks through Community Detection in Complex Networks: Tools, Trends and Challenges

Nicollas Oliveira (UFF), Dianne Medeiros (UFF), Diogo Mattos (UFF)

The echo chamber is a phenomenon related to the tendency of users of social networks to interact with other users in homogeneous groups and with similar ideas and opinions. As a result, the echo chamber harms the contradictory and encourages the phenomenon of confirmation bias, fostering environments conducive to hate speech and the spread of fake news. The short course presents the main algorithms for structural characterization and techniques that help in the detection of echo chambers. The minicourse focuses on community discovery approaches on a topology graph created according to the diffusion of information in social networks. Algorithms for characterizing complex networks and the performance indices of these approaches are also detailed. In addition, the mini-course develops a practical activity of capturing data on social networks and analysis to identify echo chambers. Finally, the challenges and research projects that focus on the study of echo chambers in online social networks are discussed.

Generation of Time Series Using Adversarial Generative Networks: From Theory to Practice

Iran Ribeiro (UFES), Breno Krohling (UFES), Vinicius Mota (UFES), Giovanni Comarela (UFES)

Access to datasets is essential for various research fields, such as data science and machine learning. In many scenarios, however, data access and publication are limited by challenges in data collection, handling of missing information, and privacy guarantees. An alternative to tackle this problem is the generation of synthetic data based on the original data, preserving its characteristics while maintaining its privacy. In the literature, it is possible to find several models, for example, for tabular and static data. The time-series are a type of time-dependent data, it may pose additional challenges for such models. In this mini-course, we present the Generative Adversarial Networks (GANs), a framework based on deep learning for training generative models. We use an open dataset with information about bicycle rentals in cities in the United States and discuss the main concepts of time series, deep learning and GANs. In addition, we will present a hands-on with code executions and discussion of the results obtained.