A Study on IT-security Vocabulary for Domain Document Classification

The volume of published scientific literature available on Internet has been increasing exponentially. Some of them reflect the latest achievement of the specific research domain. In recent years, many projects have been funded aiming to online scientific literature mining, especially in biomedical research. Scientific literature covers most of the hot topics in the research field and has a very large domain-specific vocabulary. The exploitation of domain knowledge and specialized vocabulary can dramatically improve the result of literature text processing. In this paper, we build a large-scale annotated corpus from abstract content in IT security literature, from which we construct domain vocabulary with TF/IDF-like schema and present quantitative analysis on the difference between features constructed respectively from positive and negative - annotated corpus. We evaluated the effects of the vocabulary by document similarity computing and classifying. The experimental result shows that domain vocabulary can improve the accuracy effectively.