Enhancing Topic Modeling for User Expertise Profiling
Saida Kichou, Fouad Dahak, Abdelkrim Meziane · 2025
This paper revisits and enhances an existing approach to expertise evaluation by incorporating tag depth into the Latent Dirichlet Allocation (LDA) topic modeling process and applying the method to social bookmarking data from Delicious. By integrating social indicators (tags) along with their hierarchical depths, our approach refines topic modeling to better align identified topics with users’ actual areas of expertise, leading to more precise and relevant assessments. To evaluate this method, we analyze social tagging behavior on the Delicious platform, where users categorize and tag saved links. By leveraging both the content and hierarchical structure of these tags, the study demonstrates the effectiveness of the proposed approach in extracting focused and meaningful topics. The refined analysis not only filters relevant subjects more effectively but also provides quantifiable expertise measures, highlighting the potential of social bookmarking systems for expertise assessment and user profiling.