Pearl Pu
Emotion Analysis in Natural Language
What is human emotion? It turns out there are more than 90 definitions. Among the most recent well-accepted ones, emotion is understood as our reaction to external and internal events such as a loud noise (external, surprise), being told we passed the college entrance exam (external, joy), or a thought that triggered the memory of a bygone love (internal, sad). We care about emotions because they motivate us to take actions, influence the quality of our decisions, and enhance our ability to empathize and communicate.
In this talk, I will describe the various challenges related to understanding, detecting, and visualizing emotions in large text datasets. To show you the research methodology we have explored and developed, I will demonstrate concrete steps and systems for eliciting emotion annotations from crowd workers, motivating and incentivizing them, building domain-specific lexicons, processing modifiers such as negations and intensifiers, and detecting emotions in human dialogs such as those found in TV series and movies.
The talk will end with some suggestions for future work in this area including building emotion-aware dialog systems.
This talk is supported by ACM Distinguished Speakers Program.
Dr. Pearl Pu leads the Human Computer Interaction Group in the School of Computer and Communication Sciences at EPFL. Her research is multi-disciplinary and focuses on issues in the intersection of human computer interaction, artificial intelligence, and behavioral science. Over her long career, she has worked on decision support and recommender systems, user preference elicitation, emotion analysis in social media, and information visualization.
She serves on several editorial positions: the AI Magazine, the ACM Transactions on Interactive Intelligent Systems, the User Modeling and User Adapted Interaction, and IEEE Multimedia (past). She is a member of the steering committee of the ACM International Conference on Recommender Systems.
She holds a Ph.D. in Computer and Information Sciences from the University of Pennsylvania in the United States and was a recipient of the research initiation award (now research career award) from the US National Science Foundation, as well as 13 Research Awards from the Swiss National Science Foundation, and 3 Technology Innovation Awards. Her team won the Worldwide Innovation Challenge in recognition of her scientific and entrepreneurial contribution in emotion recognition in social media.
Jennifer Golbeck
Foretold Futures from Digital Footprints: Artificial Intelligence, Behavior Prediction, and Privacy
Personalization, recommendations, and user modeling can be powerful tools to improve people’s experiences with technology and to help them find information. However, we also know that people underestimate how much of their personal information is used by our technology and they generally do not understand how much algorithms can discover about them. Both privacy and ethical technology have issues of consent at their heart. This talk will look at how to consider issues of privacy and consent when users cannot explicitly state their preferences, The Creepy Factor, and how to balance users’ concerns with the benefits personalized technology can offer.
This talk is supported by ACM Distinguished Speakers Program.
Jennifer Golbeck is Director of the Social Intelligence Lab and an Associate Professor in the College of Information Studies at the University of Maryland, College Park.
Her research focuses on analyzing and computing with social media, focused on predicting user attributes, and using the results to design and build systems that improve the way people interact with information online.
Paul De Bra
After twenty-five years of user modeling and adaptation…what makes us UMAP?
ACM UMAP 2017 is the 25th conference on User Modeling, on Adaptive Hypermedia, or on both together (since 2009). The research has actually been going on for more than 25 years as initially there was a conference only every two years. This keynote offers both reflection on the past and outlook into the future, with the burning question: What makes us UMAP? We perform research on modeling users (individuals as well as groups), not just for fun but to use these models for recommendations and for adaptation. That’s not unique to us. In recommender systems analyzing user behavior is needed in order to give better and better recommendations, and likewise an area like educational data mining analyzes how learners study in order to best guide them to new learning material or followup courses. With analysis of social networks and website adaptation we step into the same research area that is covered by the hypertext community. If all of this is “us” but “not just us”, where is our identity?
One key characteristic of User Modeling is our quest to come up with understandable user models, or scrutable as Judy Kay coins them. The same is true for the adaptation: we strive to understand why certain adaptation happens or why a certain recommendation is given. UMAP research is not complete if we cannot understand the chain that leads from user action to (a perhaps much later) system reaction. As we move from expert-driven adaptation towards data-driven adaptation the problem of understanding the user-modeling-to-adaptation process is becoming harder and harder. But we need this understanding to ensure that adaptation continues to adapt in the right way under continuously changing circumstances (both in what we adapt and in the users and context we adapt to). We need the understanding also to prevent continuous adaptation from creating lter bubbles and to avoid creating the illusion that the recommendations will always be “right” because of the “wisdom of the crowd” principle.
One key element has always been missing from UMAP, and this keynote will ll that void: we need to practice what we preach. Therefore, the conference proceedings will only contain this abstract, but there will be a real paper to go with this abstract. That paper cannot be printed because it is adaptive.
Prof. dr. Paul De Bra was is Professor in the Department of Computer Science at the Eindhoven University of Technology (TU/e) and chair of the Web Engineering group.
Paul De Bra started his university life at the University of Antwerp, Belgium, where he first obtained a master degree in Mathematics, as well as a teaching qualification, in 1981. He continued his study towards a doctorate under the guidance of prof. dr. Jan Paredaens and graduated in 1987, with a thesis on “Horizontal Decomposition in the Relational Database Model”.
During 1988 and 1989 he was a post-doctoral researcher at AT&T Bell Laboratories in Murray Hill, New Jersey, studying principles and technology for WYSIYWYG interfaces for document processing. He also learned about the then upcoming research field of hypertext.
He joined the TU/e in December 1989, first as an associate professor in databases, and since August 1996 as a full professor and chair of the Information Systems group, out of which the Web Engineering group emerged.
Paul De Bra is especially known for research on adaptive hypermedia and adaptive Web-based systems. He initiated both theoretical research and practical development, the theory leading to the most cited reference models in adaptive hypermedia, AHAM and GAF, and the development leading to the most cited and used adaptive hypermedia systems AHA! and GALE. His largest project was GRAPPLE, an EU FP7 Technology-Enhanced Learning project to support life-long learning through the use of a user-modeling and adaptation platform that allows users to be helped by adaptation even when moving between different learning systems and providers. He also initiated the CHIP project on personalization in cultural heritage, and contributed to many other studies and development, most recently also on adaptation for autistic students moving from high-school to the university, in the Autism&Uni project.
Besides his teaching, research and grant acquisition tasks at the TU/e he served in a number of additional functions. He was a part-time professor at the University of Antwerp from 1987 to 2007, is scientific director of the dutch research school SIKS on Information and Knowledge Systems, President of User Modeling Inc., and at the TU/e he is Graduate Program Director of Computer Science and representatitive of the TU/e in the World Wide Web Consortium.
