基于KSVM的中文实体关系抽取研究-计算机应用技术专业论文.docxVIP

II Ab Abstract Title ： RESEARCH ON CHINESE ENTITY RELATION EXTRACTION BASED ON KSVM Major：Computer Application Technology Name：Miao PU Signature: Supervisor：Prof.Quanzhu YAO Signature: Abstract The main purpose of Information Extraction is to transform unstructured text for structured information, it not only can satisfy people to get valuable information from huge amounts of information demand, is also the basis of the Information Retrieval and Automatic Response System. Entity Relation Extraction is an important links in the Information Extraction, has now become a hot research topic and has broad application prospects. In recent study, people regard Entity Relation Extraction as a classification problem, construct relevant relationship instance, and use machine learning algorithm to train classifier which can determine the candidate relations belong to which kind of predefined relations. In Machine learning algorithm, the SVM (Support Vector Machine), has the highest classification accuracy, KNN(K-Nearest Neighbor) is the se. Because the SVM apply related kernel functions that can convert nonlinear classification to the linear classification problem and it has the implicit high-dimensional classification characteristics, so the entity relation extraction based on kernel function is most widely used. According to the principles of SVM and the defects of kernel function itself for the Chinese entity relation extraction, the innovation points and the main work in this paper are as follows: 1. For the SVM-based Chinese entity relation extraction method, the misclassifications are distributed to the neighbour optimal super-plane of SVM in feature space. The reason is the calculation only takes one representative point into consideration.Aiming at this problem, A Chinese entity relation extraction method based on KSVM is proposed by combining the advantages of SVM algorithm and KNN algorithm in this thesis. For sample points near the interface use KNN to classify