ISSN : 1013-0799
This study proposes a user-profiling framework for scholarly information services that integrates research stages, behavior levels, and semantic analysis to overcome the limits of keyword-based systems. The framework unifies static attributes, dynamic behavioral signals, semantic topic representations, and contextual research states to support adaptive, researcher-centered information services. Five core components structure the model: research stage modeling (SE-1), behavior-level classification (SE-2), embedding- and knowledge-graph-based semantic analysis (SE-3), log-driven dynamic profiling (SE-4), and a modular architecture for interoperability (SE-5). Structural validation confirmed that the framework satisfies ten requirements derived from prior studies, including semantic linkage, quality filtering, adaptive interfaces, and predictive recommendations. Expert evaluation by three specialists yielded an average score of 4.6 out of 5, highlighting strong theoretical grounding and practical usefulness. The main challenge identified was the scarcity of labeled data for inferring contextual user states. To address this, heuristic bootstrapping and active-learning strategies are proposed.
