The Event will be held virtually via Zoom.
Please register: https://uni-bremen.zoom.us/meeting/register/u5Ivd-qspjwsHNRXKSlr4K1xHv5GIXy31-FX
14:00-14:10 (UK) Welcome
Session 1: Chair: Lena Steinmann (Bremen University)
14:10-14:30 (UK) Remixing Political News Reception: Studying Visual Framing Processes with established and innovative methods (Stephanie Geise)
14:30-14:50 (UK) Reasoning over Incomplete Knowledge Graphs (Víctor Gutiérrez-Basulto)
14:50-15:00 (UK) Questions/Discussion
Session 2: Chair: Fernando Loizides (Cardiff University)
15:00-15:20 (UK) Trust in AI through Understanding, Control, and Co-Creation (Hendrik Heuer )
15:20-15:40 (UK) Natural Language Processing and Social Media: Challenges, Applications and TweetNLP (Jose Camacho-Collados)
15:40-15:50 (UK) Questions/Discussions
15:50-16:30 (UK) Next steps: An open discussion on collaboration in education / research and other activities.
The Event will be held virtually via Zoom.
Please register: https://uni-bremen.zoom.us/meeting/register/u5Ivd-qspjwsHNRXKSlr4K1xHv5GIXy31-FX
Cardiff University
Abstracts
Natural Language Processing and Social Media: Challenges, Applications and TweetNLP
Abstract:
Social media represents a fundamental tool to understand the interactions and progress of society in the 21st century. Despite the large amount of information generated in social media platforms, understanding what is going on is not an easy task, even after the significant NLP progress in recent years. Its multilingual, dynamic, informal and multimodal nature means that standard techniques are seldom optimal. In this talk, I will summarise some of the latest advances on social media processing, and in particular the development of specialised language models. I will present our latest project, TweetNLP (tweetnlp.org), stemming from an international collaboration between academia and industry. TweetNLP is an integrated platform that enables a simple usage of cutting-edge NLP models for social media. I will quickly showcase TweetNLP’s online demo and will discuss a few applications for social media research, in particular for analysing political communication.
Bio:
Jose Camacho-Collados is a Senior Lecturer and UKRI Future Leaders Fellow at Cardiff University, leading the Cardiff NLP group. Before joining Cardiff University, he completed his PhD in Sapienza University of Rome and was a Google AI PhD Fellow. Until recently, his research has focused on various semantics aspects in NLP with a distributional perspective. He wrote the “Embeddings in Natural Language Processing" book and is the current program co-chair of *SEM 2023. More related to the topic of the talk, in the last few years Jose has been working in social media analysis and applications, developing NLP tools specifically targeted to this domain.
Reasoning over Incomplete Knowledge Graphs
Abstract
Knowledge Graphs (KGs) are at the core of many AI applications, such as recommendation or question-answering systems. However,
KGs are inevitably incomplete and noisy. This presents a challenge for traditional deductive reasoning methods. For instance, using traditional subgraph matching methods to answer queries over incomplete KGs might lead to empty or wrong answers. In this short talk, I will briefly discuss some of our recent works using low-dimensional vector spaces to reason over incomplete KGs.
Bio
Víctor Gutiérrez-Basulto is a Senior Lecturer in the School of Computer Science and Informatics at Cardiff University and a Senior Research Scientist at the Huawei Research Lab at Edinburgh. Before that he hold two prestigious postdoctoral fellowships: Marie Skłodowska-Curie COFUND fellow within the Sêr Cymru II programme and M8 Early Career Postdoc fellow at the University of Bremen. His work focuses on Semantic Data Management and on the integration of Learning and Reasoning.
University of Bremen
Abstracts
Trust in AI through Understanding, Control, and Co-Creation
Abstract
In this talk, I will provide an overview of my efforts to achieve trust in AI through understanding, control, and co-creation. My ultimate goal is to empower users to shape their interactions with AI and social media platforms. A special focus lies on using AI to help users find reliable information and understand complex text. To build such platforms, I have developed methods to measure and analyze users' understanding of AI systems, as well as methods to control the output of AI systems through audits. I have also explored how explanations of AI systems can be guided by the understanding of users. I will explain why involving users in the co-creation of AI systems is crucial and how this can be used to fight misinformation.
Bio
Dr. Hendrik Heuer is a postdoctoral researcher at the Institute of Information Management Bremen (ifib) and the Centre for Media, Communication, and Information Research at the University of Bremen. His research focuses on human-computer interaction and machine learning. He is currently working on ways to combat misinformation. He studied and worked in Bremen, Stockholm, Helsinki, and Amsterdam and was a Visiting Postdoctoral Research Fellow at Harvard University.
Remixing Political News Reception: Studying Visual Framing Processes with established and innovative methods
How do media images in multimodal media environments influence the reception, processing and interpretation of political information? This question is investigated using the example of online news in which visual and textual information (in the form of news texts and press photos) are combined to cover politics. Based on theories of reception and processing of visual vs. textual information and the concept of framing, we examine the sensory perception of visual political information in its multimodal context and focus on related affective and cognitive media effects. For the empirical analysis, we implemented an innovative multi-method design that combined traditional post-receptive (survey; guided interviews) and computational, continuous measurements (eye tracking, automated emotion detection) to systematically analyze the multimodal reception process and its effects on issue believes and political participation. The study promises to provide important new insights into the reception and impact of multimodal political media information. In our talk, we focus on the challenges and potentials arising from the complex interweaving of computational observational data with survey items.
Bio:
Stephanie Geise studies how people perceive and process political media content through images and texts and how multimodal media messages - for example, traditional news articles, election advertising, or fake news online - affect political thinking and action. In various research projects, she has combined survey methods with innovative, computational tracking methods (e.g., eye tracking, automated emotion recognition). Since April 2021, she is filling in for the professorship of Communication and Media Studies with a focus on digital communication.