Research On The Method Of Extracting Entity 's Attribute Relationship

The entity attribute relation extraction of unstructured free text, is important foundation for constructing ontology knowledge base and information extraction. Therefore, to study the methods that how to extract the entity attribute relation, is a very meaningful work. This paper introduces the research on entity attribute relation extraction,some work has been done as follows:(1) This paper presents a novel method based on LM algorithm to extract entity attribute relation. This method takes the relation extraction problem as multi-classification problem. Constructing and training the relation classifier with the advantages of powerful nonlinear reflection and the self-learning ability of BP neural network, the global optimization and fast convergence speed of LM algorithm. And finally, the entity attribute relations are extracted through the relation classifier by this method. Compared with the traditional method based on SVM plus inference rules, this method improves the performance of attribute relation extraction.(2) In order to accelerate the convergence speed of the BP neural network and improve the performance of relation extraction, this paper presents a optimization method based on the combination of particle swarm optimization algorithm and LM algorithm. This method uses PSO algorithm to optimize connected weights values of BP neural network, then uses LM algorithm to adjust connected weights values. It can overcome the disadvantages in BP neural network, such as slow convergence rate, long training time and so on. Experiments show that this kind of improved algorithm based on the combination of LM algorithm and PSO algorithm has better feasibility and practicality in relation extraction.(3) As a new machine learning algorithm, deep belief networks (DBN) model can automatically learn the combined features that are much more conducive to classify. DBN is composed of several unsupervised RBM networks and a supervised BP network. The RBM layers learn combination features automatically, maintain as much information as possible when feature vectors are transferred to next layer. The BP layer trains relation classifier depending on the combination features learned by the last RBM layer, and fine-tune the whole networks. Compared to the methods of (1) and (2),this method is more suitable for the information extraction task of high-dimensional space features. And it has better classification effect.Experimental results show that the above methods can improve the performance of entity attribute relation extraction. These researches established good foundation for building tourist ontology knowledge base.