From preference to purpose: a reconfigurable recommendation framework for user-centred information access

This study aims to develop a reconfigurable recommendation framework for academic information services that dynamically adapts to users’ changing information seeking purposes.

Grounded in information behaviour theories, a multi-aspect recommendation framework was designed by profiling user behaviour and content attributes. The system was implemented as a prototype on an actual academic platform and evaluated through a preliminary, proof-of-concept user study involving 66 participants.

The evaluation demonstrated statistically significant improvements in users’ search experience and perceived utility. Furthermore, large-to-very-large effect sizes (Cohen’s d) indicated that these improvements were not only statistically significant but also practically meaningful. The framework was positively received across all user groups.

The evaluation is a preliminary, user-centred study (N = 66) focused on perceived utility and search experience, rather than quantitative performance metrics like precision or recall. It establishes the potential of the framework in a prototype environment. Future work should validate these findings through large-scale, longitudinal studies with baseline comparisons and incorporate more diverse contextual variables to enhance the model’s robustness and generalizability.

The framework enables flexible application in academic environments by offering differentiated strategies based on user intent and behaviour. It provides practical guidance for designing explainable, adaptive information services.

While prior studies have explored multi-aspect models, they often rely on static configurations. This study presents a theoretically grounded, reconfigurable framework where the entire recommendation pipeline – from content evaluation criteria to strategy selection – dynamically adapts to evolving user intent. By integrating user profiling based on engagement levels with a flexible multi-aspect content structure, our framework offers a novel approach to structural flexibility in recommender system design, moving beyond fixed algorithmic optimizations to achieve purpose-driven personalization.