One of the most critical activities of revealing terrorism-related information is classifying online documents.The internet provides consumers with a variety of useful knowledge, and the volume of web material is increasingly growing. This makes finding potentially hazardous records incredibly difficult. To define the contents, merely extracting keywords from records is inadequate. Many methods have been studied so far to develop automatic document classification systems, they are mainly computational and knowledge-based approaches. due to the complexities of natural languages, these approaches do not provide sufficient results. To fix this shortcoming, we given approach of structure dependent on the WordNet hierarchy and the frequency of n-gram data that employs word similarity. Using four different queries terms from four different regions, this approach was checked for the NY Times articles that were sampled. Our suggested approach successfully removes background words and phrases from the document recognizes connected to terrorism texts, according to experimental findings.