Talk 1:  Exploring the Frontiers of Antisocial Detection and Analysis Through Deep Learning

Abstract: The use of online platforms has become increasingly integrated into people's daily lives. In this context, antisocial online behaviour has become a pervasive issue, with a wide range of harmful actions (e.g., cyberbullying, hate speech, online harassment, misinformation, and disinformation) occurring on various platforms.

This presentation delves into the critical challenges we're tackling in our ongoing research on online antisocial behaviour, particularly focusing on analysing, detecting, and mitigating hate speech, misinformation, and disinformation. The presentation will include topics such as (i) innovative deep learning architectures designed to identify various forms of antisocial behaviour, (ii) the importance of incorporating both contextual factors (e.g., user interactions and reactions) and content information from online posts to train effective models, and (iii) the potential benefits of extracting polarity from messages and posts to enhance detection accuracy. The presentation will also explore the application of topic modelling techniques to extract underlying themes from malicious messages and categorize them based on their type and geographic origins.

 Elena Simona Apostol is an Associate Professor of Computer Science at the Faculty of Automatic Control and Computers, National University of Science and Technology Politehnica Bucharest. In 2024, she received her habilitation in computer science with a focus on natural language processing and deep learning techniques. Throughout her career, she has contributed to research at various institutions and universities, including the Institute for Research in Digital Science and Technology (INRIA) France, Microsoft Research Center, Fraunhofer FOKUS - NGNI department, and Uppsala University. With over 100 peer-reviewed scientific publications and an i10-index of 29 and h-index of 16, she has made scientific contributions in the fields of Data Science, Machine Learning, Deep Learning, Natural Language Processing, Text Mining, Data Mining, Big Data Analytics, Time Series Analysis, and Cloud Computing. Her current research focuses on the effects of disinformation, misinformation, and harmful speech in social media, as well as novel and effective methods for detecting and mitigating its spread throughout online communities.

Talk 2: Challenges of Hate Speech Detection in Online Communication Among Adolescents

Abstract: This presentation will explore key challenges we are addressing in our current research on hate speech detection in the context of adolescent online communication. These challenges include variations in taxonomies of annotated datasets, semantic differences in hate language, and biases within detection algorithms. We will showcase collected datasets, introduce a youth language bias detection framework we have developed, and highlight practical applications within educational settings.

In 2019, Jan Fillies received his Master’s degree in Information Systems from the Technical University of Munich. His master thesis focused on AI-based pattern identification. He is pursuing a Doctorate degree in Computer Science from the Freie Universität Berlin, researching the influence of youth language on algorithmic hate speech detection. During this process, he is collecting data, developing prototypes for educational outreach in schools, and training multilingual detection models. Utilizing symbolic AI methods to combine the sparsely available datasets and enhance the range of definitions of hate covered by the trained networks.

Talk 3: What's next after Antisocial Behaviour Detection? Large Graph Network Immunization

Abstract: Network immunization is an automated task in the network analysis field that involves protecting a network (modeled as a graph) from being infected by an undesired arbitrary diffusion. In this talk, we focus on three types of network immunization approaches: counteractive approaches (e.g., SparseShield), tree-based approaches (e.g., MCWDST), and community-based approaches (e.g., CONTAIN). For the counteractive approach, we will present SparseShield which determines the best node to immunize after the network is infected. MCWDST is a tree-based approach that constructs the minimum-cost weighted directed spanning tree for a harmful node to determine the best nodes to immunize. Finally, CONTAIN, a community-based network immunization approach, generates partitions and ranks them for immunization using the subgraphs induced by each harmful node.

Ciprian-Octavian TRUICĂ holds an Assistant Professor of Computer Science position at the Computer Science and Engineering Department, Faculty of Automatic Control and Computers, National University of Science and Technology Politehnica Bucharest. In 2024, he defended his habilitation thesis in Computer Science on the topic of Textual Data Management and Analysis using Big Data Analytics and Natural Language Processing. In 2018, he received his Ph.D. degree in Computer Science and Information Technology on the topic of data management and text mining from the University Politehnica of Bucharest, Romania. He held research and teaching positions at Uppsala University (Sweden), Aarhus University (Denmark), and Université de Lyon (France). With more than 85 peer-reviewed scientific articles, he has made substantial scientific contributions in the fields of Natural Language Processing, Machine Learning, Deep Learning, Computational Linguistics, Data Science, Big Data Analytics, and Data Management.