Exploiting various information for knowledge element relation recognition

Knowledge element relation recognition is to mine intrinsic and hidden relations, i.e., preorder, analogy and illustration from knowledge element set, which can be used in knowledge organization and knowledge navigation system. This paper focuses on what information is employed to recognize knowledge element relations. First, a formal definition of knowledge element and the types of relation are given. Next, an algorithm for knowledge element sort is proposed to gain the sequence number of knowledge element. Then, information of term, type, distance, knowledge element relation level and document level is selected to represent candidate relation instances. Evaluation on the four data sets related to "computer" discipline, using Support Vector Machines, shows that term, type and distance features contribute to most of the performance improvement, and incorporation of all features can achieve excellent performance of relation recognition, whose F1 Micro-averaged measure is above 83%.