How Many Keywords are Enough? Determining the Optimal Top-K for Educational Website Classification

The classification of educational websites has become increasingly challenging, as traditional indicators such as domain extensions no longer reliably reflect a site’s purpose. This study investigates the optimal number of TF-IDF-ranked keywords (K) required to balance classification accuracy and computational efficiency in a one-class setting. Using a curated dataset sourced from the DMOZ directory and verified educational websites, multiple Top-K keyword subsets (K = 10–200) were evaluated. A One-Class Support Vector Machine (SVM) was employed, with performance assessed through cross-validation and separate positive/negative test sets. Results indicate that classification accuracy peaks within the range of K = 30–100, with diminished performance beyond this range due to the inclusion of irrelevant or noisy terms. These findings offer a practical and scalable framework for content-based educational website classification, particularly for applications in low-resource environments, and challenge the default reliance on exhaustive keyword feature sets.