Program

The AltRecSys workshop will be held on Friday, October 18, 14:15-17:45 in Room L at the Politecnico di Bari.

  • 14:15-14:25: Chairs’ Welcome
  • 14:25-15:00 Highlight Talk: Post-Userist Recommender Systems: A Manifesto - Robin Burke
  • 15:00-15:45 Discussion
  • 15:45-16:15 Break
  • 16:15 -16:30 Lightning talks
    • Kim Falk & Robin Verachtert: Let’s Talk about Sex, Mr ChatBot
    • Lucas Vinh Tran & Jay Katukuri: A Centralized Configuration Driven Modeling Framework for Personalization in Banking and Finance
    • Bart P. Knijnenburg & Sushmita Khan: Let Me Show You You: Personalized Preference Profiles for Self-Actualization
    • Florian Atzenhofer-Baumgartner, Bernhard Geiger, Christoph Trattner, Georg Vogeler, & Dominik Kowald: Challenges in Implementing a Recommender System for Historical Research in the Humanities
  • 16:30-17:30 Discussion
  • 17:30-17:45 Wrapup

Details for the (Lighting) Talks

  • Post-Userist Recommender Systems: A Manifesto – Robin Burke, Information Science, University of Colorado, Boulder; Morgan Sylvester, Independent Scholar, Louisville, Colorado

Abstract: We define userist recommendation as an approach to recommender systems framed in terms of the relation between the user and system. Post-userist recommendation posits a larger field of relations in which multiple stakeholders are embedded and distinguishes a recommendation function (which can potentially connect creators with audiences) from a generative one. We argue that in the era of generative media, userist recommendation becomes indistinguishable from personalized media generation, and therefore post-userist recommendation is the only path forward for recommender systems research. Extended Version

  • Let’s Talk about Sex, Mr ChatBot – Kim Falk, Binary Vikings & Robin Verachtert, Binary Vikings

Abstract: Almost any news source, whether a regular newspaper, LinkedIn, or similar, will tell you these days that ChatGPT and LLMs will generally solve any problem, steal your job and do a better job. We see one blog post after another showing how to create a cold start recommenders with an LLM without providing user data that outperforms SOTA algorithms. The LLM just “knows.” However, this is only true for domains and topics often discussed on the internet and topics falling within the ethics and rules of the country where the model was created or indeed, the company itself. As soon as you move away from those, you start having problems. Sex toys and Scandinavian languages might both be extreme examples, but it does illustrate problems and biases which are introduced into our systems. Extended Version

  • A Centralized Configuration Driven Modeling Framework for Personalization in Banking and Finance – Lucas Vinh Tran, JP Morgan Chase & Jay Katukuri, JP Morgan Chase

Abstract: In this talk, we will discuss a centralized AI/ML modeling framework uniquely engineered for the domain of personalization and recommender systems in the banking and finance sector. Departing from conventional methodologies, this framework provides basic building blocks and tools for streamlined model development and governance, allowing for rapid experimentation and implementation through minimal configurations. The framework fosters collaboration among diverse teams, ensuring coherence, transparency, and adherence to governance, legal, and compliance protocols throughout the model development pipeline. This speaking engagement in the workshop will provide the audience with general information to recognize challenges and navigate common pitfalls when building personalization models in the banking and finance sector.

  • Let Me Show You You: Personalized Preference Profiles for Self-Actualization – Bart Knijnenburg, School of Computing, Clemson University & Sushmita Khan, School of Computing, Clemson University

Abstract: Recommender systems learn users’ preferences by tracking their consumption or purchase behaviors, but this functionality assumes that users have a general idea of what their preferences are. Research in consumer behavior shows that this is not always the case–users tend to mostly have myopic preferences that are heuristically constructed on-the-fly. We propose Personalized Preference Profiles (PPPs) that help users reflect on their preferences in a way that meaningfully supports their preference construction process. PPPs visualize users’ preferences rather than the recommendation process, which allows users to realign their preferences with their long-term goals. In our talk we aim to discuss various considerations for creating effective PPPs. Extended Version

  • Challenges in Implementing a Recommender System for Historical Research in the Humanities – Florian Atzenhofer-Baumgartner, Institute of Interactive Systems and Data Science, Graz University of Technology, Graz, Austria Bernhard C. Geiger, Know Center Research GmbH, Graz, Austria, Christoph Trattner, University of Bergen/Media City Bergen, Bergen, Norway, Georg Vogeler Department of Digital Humanities, University of Graz, Graz, Austria, & Dominik Kowald, Institute of Interactive Systems and Data Science, Graz University of Technology, Graz, Austria

Abstract: Recommender systems (RecSys) have become indispensable tools in various domains. However, applying RecSys within humanities and historical research presents unique challenges. Unlike conventional implementations focused on consumer goods or media, RecSys in these fields must navigate the complexities of cultural records, such as historical legal documents, e.g., charters. Three main points highlight why this needs special consideration: Firstly, unique item characteristics: Charters are distinct due to their varied, high-dimensional metadata, public value as cultural heritage, and need for context-sensitive interpretations. Secondly, distinctive user behavior: Historians’ information-seeking behavior is crucial for RecSys design in this domain. Thirdly, multi-stakeholder complexity: Implementing RecSys involves navigating interests of multiple stakeholders with distinct and sometimes conflicting goals. In summary, deploying effective RecSys in humanities and historical research demands tackling three major challenges: addressing historical document facets, distinctive scholarly user behaviors, and conflicting goals in a non-conventional multi-stakeholder setting. Success could significantly enhance historical scholarship and cultural heritage curation, and deepen our understanding of the past. Extended Version