Assessing the performance of machine learning techniques in achieving keyword coverage through semantic data analysis

Abstract

This article provides an in-depth comparative analysis of two advanced hybrid machine learning models designed for keyword extraction. The first combines Bidirectional Encoder Representations from Transformers (BERT) with an Autoencoder (AE), while the second merges the traditional Term Frequency–Inverse Document Frequency (TF-IDF) method with the same autoencoding framework. The study focuses on semantic analysis within text corpora, aiming to assess how effectively these methods identify meaningful and comprehensive keyword sets across various datasets. A thorough examination of each model’s architecture and mechanisms is performed, emphasising the crucial role of autoencoders in maintaining semantic consistency and contextual relevance of the extracted keywords. The experimental analysis covers multiple datasets, illustrating how differences in textual structure and semantic richness impact the overall extraction performance. Performance metrics such as precision, recall, and the F1-score are utilised to evaluate and compare accuracy. Additionally, the discussion outlines the respective strengths, limitations, and ideal scenarios for applying both hybrid approaches. Ultimately, the findings offer valuable insights for researchers and practitioners, providing practical guidance on selecting the most suitable method for tasks requiring deep semantic understanding and high precision in information retrieval.