A Multilevel N-gram Model with Naïve Bayes Classification of Personal Web History Datasets

Abstract

The rapid expansion of 4G and 5G networks has accelerated the proliferation of Internet-connected devices, leading to a massive increase in Internet of Things (IoT) traffic and the generation of diverse big data. Big data analytics has been widely adopted across healthcare, gaming, cybersecurity, business, and other sectors to extract actionable insights, uncover hidden patterns, and support informed decision-making on their targeted datasets. However, big data analytics of personal history datasets is a passive and underrated field. This paper addressed the gap by proposing a novel multilevel N-gram model combined with a Naïve Bayes classifier to classify the personal website history datasets. In the first stage, URL strings are decomposed into multiple N-gram levels (unigrams, bigrams, and trigrams) to capture both simple lexical features and contextual patterns. In the second stage, the extracted features are classified using the Naïve Bayes algorithm, which applies Bayes’ theorem under the assumption of conditional independence to compute category probabilities. Empirical validation on a standardised dataset demonstrates that the proposed approach achieves an average F1-score of 88%, outperforming existing baseline methods documented in prior literature. These findings highlight the effectiveness of the proposed method for big data analysis of web usage, particularly for personal history datasets