Skip to Main Content
Automatic text classification for a Web collection is a challenge task, especially in the case that the language is not English, such as Thai. However, most of Thai educational Web pages usually include English terms due to their technical aspect. Lots of technical terms and typing errors both in Thai and in English are found in Web sites of universities. Most previous works on text categorization applied term frequency and inverse document frequency for representing importance of terms. In this paper, we use inverse class frequency instead of inverse document frequency in centroid-based text categorization because it works well on a collection with a large number of unique terms. The experimental results show that inverse class frequency is useful, especially when it is applied on both prototype and query vectors.