Publikation
Enhancing Entity Recommendation for Personal Information Assistance Using LLM-based Adaptive Relevance Prediction
Mahta Bakhshizadeh; Heiko Maus; Andreas Dengel
In: Information Management. International Conference on Information Management (ICIM-2025), located at 11th International Conference on Information Management, March 28-30, London, United Kingdom, Springer CCIS series Conference Proceedings, 3/2025.
Zusammenfassung
Recent studies in the domain of Recommender Systems have increasingly leveraged Large Language Models (LLMs) to advance various objectives, including improving recommendation accuracy, enhancing user experience, and providing more transparent explanations. Continuing along this path, we apply LLMs in a relatively novel direction: Personal Information Assistance, with a focus on Entity Recommendation to support Knowledge Workers. Our approach utilizes the Mistral 7B model to perform adaptive relevance prediction, effectively incorporating users' explicit feedback to refine recommendations over time. By adapting to user preferences, our method aims to deliver more precise and contextually relevant recommendations. We evaluated our approach using a benchmark built upon RLKWiC, a dataset of Real-Life Knowledge Work in Context. The results show that our LLM-based method significantly outperforms both the baseline approach and a recently proposed semantic-based method for Entity Recommendation, demonstrating the potential of LLMs to further advance the capabilities of Personal Information Assistance systems.
Projekte
- Managed_Forgetting_Phase_2 - Nachhaltige evolutionäre Unternehmensgedächtnisse: Methoden und Effekte von Managed Forgetting für die administrative Wissensarbeit
- SensAI - Self-organizing Personal Knowledge Assistants in Evolving Corporate Memories