A Comparative Analysis of Sentence Embedding Techniques and LSTM Models in Web Page Classification

Given the growing volume of online content, web page classification is crucial for ensuring accurate and efficient information retrieval. In this study, we explore several sentence embedding techniques, based on transfer learning models, in the context of a web page classification task. We further compare three of these sentence embedding techniques, with three other LSTM architectures: a vanilla LSTM model, bidirectional LSTM (BiLSTM) and LSTM with attention mechanism. The dataset we used contains roughly 11000 web pages that have been legally scraped from a well known news website. The comparison shows that BERT sentence embedding outperforms all models, including the LSTM hybrid model with attention mechanism.