A Novel Method for News Recommendation on Websites Using the Clustered-Vectors Optimization Algorithm

Recommendation systems have been extensively implemented across various sectors, including e-commerce, social media, and banking. Numerous studies have employed established machine learning techniques to enhance the performance of recommendation models to a certain degree. This research introduces a recommendation model for news content utilizing a novel machine learning technique known as the “Clustered-Vectors Optimization (CVO)” algorithm. The proposed algorithm aims to optimize the clustering of news titles to improve the performance of news recommendations. Once the news titles are effectively clustered, a recommendation list is generated for website visitors based on their experiences. The recommendation list for each visitor is selected from news titles within the same cluster that previous visitors have accessed. This study utilized two sources of news datasets. The primary dataset, collected from a back-end website, was provided by a private digital television company in Thailand, while additional datasets were sourced from publicly available data on Kaggle. Experimental results indicated that the proposed CVO algorithm outperformed five well-known algorithms, which were TF-IDF, Word2Vec, Doc2Vec, Bag-of-Words (BoW), and Transformer, in terms of predictive performance. For instance, on the Thai news dataset, the CVO algorithm achieved an accuracy of 97.56%, a precision of 94.59%, a recall of 97.36%, and an F1-score of 97.46%. Similarly, the English language dataset, such as Indian news, also demonstrated high performance, with the CVO algorithm achieving an accuracy of 99.90%, a precision of 99.41%, a recall of 99.62%, and an F1-score of 99.53%.