Intelligence and Security Informatics

Information extraction (IE) is of great importance in many applications including web intelligence, search engines, text understanding, etc. To extract information from text documents, most IE systems rely on a set of extraction patterns. Each extraction pattern is defined based on the syntactic and/or semantic constraints on the positions of desired entities within natural language sentences. The IE systems also provide a set of pattern templates that determines the kind of syntactic and semantic constraints to be considered. In this paper, we argue that such pattern templates restricts the kind of extraction patterns that can be learned by IE systems. To allow a wider range of context information to be considered in learning extraction patterns, we first propose to model the content and context information of a candidate entity to be extracted as a set of features. A classification model is then built for each category of entities using Support Vector Machines (SVM). We have conducted IE experiments to evaluate our proposed method on a text collection in the terrorism domain. From the preliminary experimental results, we conclude that our proposed method can deliver reasonable accuracies.