Data Science goes Tampere: iSchool Special
Speaker 1
Adam Roegiest (VP of Research and Technology at Zuva)
Title
Rethinking IR and Users in the Context of Contract Analysis
Abstract
Generative AI (and Generative IR) have caused researchers and practitioners to step back from existing notions of how users find, consume, and use information. At the same time, many of the adaptations still focus on a particular part of a user's journey (e.g., search and RAG) rather than examining a user's underlying goals and objectives and whether the existing dominant approach is still applicable and appropriate. In this talk, we will look at how we can use this moment to step back and re-examine a user's process for finding information in the context of contract analysis with ties to broader themes that are applicable to more general research endeavours. In particular, we will explore how moving beyond tackling problems at face-value can unlock new insights into how information is accessed and used.
Short Bio
Adam Roegiest is the VP of Research and Technology at Zuva, a company which offers contract analysis software in API and UI applications, and was previously the Director of Research at Kira, one of the leading purveyors of contract analysis software for law firms, and received his PhD in Legal Information Retrieval from the University of Waterloo under the supervision of Gordon V. Cormack. Adam also coordinated Text Retrieval Conference tracks and was an organizer for the Search Futures Workshop at ECIR 2025.
___________________
Speaker 2
Jussi Karlgren (Industrial researcher, Docent at University of Helsinki)
Title
Cultural Robustness and Generative AI
Abstract
Generative language models are able to produce well formulated text and participate reasonably fluently in conversation. The linguistic competence is acquired through training the model on very large data sets of language; the conversational competence through instruction training, domain-specific fine tuning, and alignment with relevant behavioural conventions. The latter - post-training - actions are in effect creating a voice out of general language. The post training data sets are specific to a language but are not specific to a cultural area: most often they are translations of sets originally made for the language the model first is trained in. We are creating test suites to assess the cultural fit, robustness, and diversity of generative language models, and this talk will give examples of such experimentation. (Similar lines of argument can be made for visual models.)
Short Bio
Jussi Karlgren is a mathematically oriented linguist with a professional background in information retrieval and human computer interaction. He is currently a principal AI scientist at AMD Silo AI, where he works on quality assessment of generative language models.
___________________
Speaker 3
Sanna Kumpulainen (Professor, Tampere University)
Title
Situated intelligence: Understanding Real-World Information Access
Abstract
Generative AI has prompted a rethinking of how people access and engage with information. While much attention has been given to improving specific components of human information interaction like searching, less focus has been placed on understanding users’ broader goals, contexts, and underlying decision-making processes. This talk explores how we can reconsider the information seeking process—from intent to interpretation and beyond—through examining real-world practices. By combining behavioural data with qualitative insights, we can begin to uncover how generative systems might better support complex knowledge work.
Short Bio
Sanna Kumpulainen is Professor of Information Studies at Tampere University, Finland. Her research focuses on interactive information retrieval and task-based information searching, with particular interest in how people engage with information systems in real-world contexts. She works at the intersection of human information behaviour and information retrieval system evaluation, contributing to both theoretical understanding and practical applications in knowledge-intensive environments.
Registration
Participation is free. Please regisister here