Machine learning and rule-based embedding techniques for classifying text documents

Abstract Rapid expansion of electronic document archives and the proliferation of online information have made it incredibly difficult to categorize text documents. Classification helps in information retrieval from a conceptual framework. This study addresses the challenge of efficiently categorizing text documents amidst the vast electronic document landscape. Employing machine learning models and a novel document categorization method, W2vRule, we compare its performance with traditional methods. Emphasizing the importance of tuning hyperparameters for optimal performance, the research recommends the W2vRule, a word-to-vector rule-based framework, for improved association-based text classification. The study used the Reuters Newswire dataset. Findings show that W2vRule and machine learning can effectively tell apart important categories. Rule-based approaches perform better than Naive Bayes, BayesNet, Decision Tables, and others in terms of performance metrics.