A Hybrid Ensemble Approach Integrating Machine Learning and Deep Learning with Sentence Embeddings for Webpage Content Classification

With the continuous increase in online information, the automatic classification of diverse and complex webpage content has become essential. In our study, we propose a novel ensemble learning method that incorporates six different sentence embedding techniques, which are trained using both traditional and deep learning models. Our method utilizes a stacking technique to calculate prediction probabilities. To test the efficacy of our approach, we compare it with two well-known ensemble methods: XGBoost and Random Forest. The comparison includes testing the models on each sentence embedding separately, and jointly through the combination of all embeddings in one dataset. We also compare our method with ensemble techniques like probability averaging and majority voting on both XGBoost and Random Forest. Our method obtained an improvement in classification metrics when compared to both XGBoost and Random Forest, on a scraped dataset containing roughly 15,647 labeled web pages.