Comparative analysis of keyword-based and semantic search models for Vietnamese online news data

Abstract

A common activity in web browsing involves retrieving webpage content based on specified keywords, such as querying pages through widely used search engines like Google. In this context, text retrieval is understood as the systematic process of identifying, filtering, and selecting documents that are most relevant to a user’s information need or query. This article focuses on the processing of Vietnamese textual data collected from online sources. The texts undergo linguistic preprocessing steps to ensure they are suitable for both keyword-based and semantic search tasks. Several established retrieval models are examined and compared, including the Boolean (BOOL) model, TF-IDF, Word2Vec, and BERT. Following this comparative analysis, the study further explores the BM25 ranking model as well as hybrid search approaches that integrate multiple techniques. The dataset utilized in this research consists of articles from the VnExpress online newspaper, specifically collected from January 2026. Experimental evaluations are conducted in the Google Colab environment using Python programming tools. Based on the obtained results, the study recommends adopting a hybrid search model that combines BM25 with advanced language models such as BERT or PhoBERT in order to significantly enhance overall search effectiveness and retrieval performance.