Research On Computational Intelligence Based Self-Adaptive Semantic Network System

As the knowledge representation methods simulating human language, semantic network has significant impact and position in intelligence system development and machine learning fields. Modern intelligence decision system needs to solve the problem how to acquire knowledge and how to refine knowledge besides represent and store knowledge. Hence, this paper mainly focus on how to introduce knowledge acquirement and knowledge refinement into semantic network system in order to make it more applicable in intelligence decision system development.In many applicable project constructed by semantic network, the system not only need to represent knowledge, but also should solve how to acquire knowledge and the refinement mechanism, which contribute the footstone for system intelligence and the crucial path to machine learning and evolution. The system which can have the acquirement and refinement mechanism can be regarded as advance intelligent system. However, most present research work pay more attention on the express the knowledge in some fields by semantic network, rather than the other aspects of semantic network. In fact, the knowledge acquirement and refinement mechanism shouldn't be separated with the representation, which is figured out by many literatures in different area. Due to these, the paper establishes the general framework to integrate the knowledge representation, acquirement and refinement in order to enhance its value in application.Since the semantic network is a very flexible knowledge representation methods, the past work mainly deal with the notion model and have less applicable. According to the process that people recognize things and learn the law and after analyzing the existing semantic network study and simplify approaches, such as semantic graphic analysis model or semantic network predigesting set, this article constructs more applicable self-adaptive semantic network formalization model. The model divide the knowledge in semantic network into different levels, from simple to complex, from tacit to explicit, from lowness to advance and so on. It not only emphasize the track and level by which knowledge flow and develop, but also make the later work about semantic network learning and refinement more strategy and directly without loss of generalization.To solve the knowledge study problem in self-adaptive semantic network system, the problem import support vector machine to set up the relationship between different level knowledge and establish self-adaptive semantic network system knowledge acquirement mechanism. Furthermore, the author makes appropriate the modifications for classic support vector machine according to the character of semantic network, which consist of: 1) define the character function in semantic space and infer the corresponding kernel function. In addition, a dynamic program algorithm is given to speed up the calculator. 2) Import soft margin optimization into support vector machine to eliminate the train error caused by the sample conflict. 3) Discuss the multi-value problem in the semantic network system study and compare the different methods' impact on the problem. These methods are composed of support vector machine combination model and support vector machine regression model. 4) Import the Bayes method into the support vector machine kernel function update procedure, which can make the kernel function adapt by the sample persistently.In the refinement fields, the paper apply the rough set approaches to solve this problem by a series operation consisting of decision table generation, decision table refinement and system reverting. Main works here are: 1) discuss how to transfer the knowledge in semantic network into decision tables and how to deal with the imperfection data in the initial table. 2) Apply the Iterative Deepening A* algorithm (IDA*) to simplify the decision table and combine the target function in classical approaches into heuristic function of Iterative Deepening A* algorithm. The train result is apparently outstanding.The paper uses two concrete applications: self-adaptive semantic tongue diagnosis system and enterprise semantic description system to explain the theory and methods are feasible and applicable. By a lot of experiments, author demonstrate how to divide the level, select the notion for different level, establish the mapping relation of these levels and set some parameters, which will contribute a lot to practice.In summary, the paper make a lot of theory and experiment research and take part in project of self-adaptive semantic tongue diagnosis system and data description, discovery, integration center(DDDIC). The research will make knowledge acquirement and refinement more applicable in semantic network system. This paper studies a lot both on theory and experiments, the research accords with trends of intelligent system development, which has great significance on actual practice.