| The Use of Model Clustering for Knowledge-Based Web Browsing Behaviour Prediction | |
|
MSRDG International Journal of Computer Scientific Technology & Electronics Engineering
© 2025 by MSRDG IJCSTEE Journal Volume 1 Issue 2
Year of Publication: 2025 |
Paper Download Article ID MSRDG-IJCSTEE-V1I2P102 |
|
Abstract: Understanding and accurately predicting web browsing behavior is a foundational challenge in personalized information retrieval and adaptive web systems. Existing approaches typically rely on single-model paradigms that fail to capture the diverse, heterogeneous nature of user navigation patterns. This article proposes a novel framework, termed Knowledge-Based Model Clustering Ensemble (KB-MCE), which integrates a domain-specific knowledge base with an ensemble of unsupervised clustering algorithms—K-Means, DBSCAN, and hierarchical agglomerative clustering—to model multi-faceted browsing behavior. The knowledge base encodes semantic ontologies and contextual rules that guide feature enrichment and cluster interpretation. A weighted ensemble fusion mechanism combines the outputs of individual clustering models, yielding coherent behavioral profiles subsequently leveraged for next-page prediction, dwell-time estimation, and interest classification. Extensive experiments on four benchmark datasets (MSNBC, CTI, BMS-POS, and a synthesized knowledge-log corpus) demonstrate that KB-MCE achieves an accuracy of 91.2%, a macro F1-score of 90.1%, and a recall of 89.8%, outperforming seven competing methods including CNN-LSTM and Random Forest baselines by margins of up to 8.5 percentage points. Scalability analysis confirms near-linear growth in training time up to 250,000 sessions. These results substantiate the effectiveness of knowledge-guided cluster-of-models strategies for behavioral prediction in dynamic web environments. |
|
| Keywords: Web browsing behavior · Knowledge base · Clustering of models · Ensemble learning · User behavior prediction · Session mining · Personalization | |
