Research On Several Key Techniques In Knowledge Fusion

Knowledge fusion is an intersectant research subject between knowledge engineering/science and information fusion. It can organize distributed homogeneous and heterogeneous knowledge resources, transfer and integrate those knowledge elements to generate new useful ones. It can also optimize their structure and provide knowledge based service during the fusion process. Study on knowledge fusion is helpful for sharing knowledge in distributed environments, cooperating among knowledge systems and optimizing quality of knowledge based service. What's more, it is valuable for research on knowledge discovery in knowledge and how to create, organize, evaluate and optimize new knowledge objects.Presently research on knowledge fusion is dispersive, so it is formalized and systemized in this dissertation. Several key modules in knowledge fusion framework are studied such as formal description of knowledge objects, fusion algorithm, structure of fused results and feedback-adjust mechanism.On analyzing traditional methods, a knowledge fusion framework based on ontology and meta knowledge is proposed. The logic model and the least function of a general knowledge fusion system are also given in a formal way.Fusion algorithm is one of the most important aspects of knowledge fusion, but the amount of effective ones is very less. Based on the typical genetic algorithm, here a new knowledge fusion algorithm is presented as KFA-G Meanwhile another new one using knowledge objects' fusing constraints is also provided as KFA-R. Cooperating with each other and emphasizing particularly on different effects, KFA-GR and KFA-RG are proposed in succession. To adapt requirement-driven environment, a new fusion algorithm KFA-SO is also proposed finally. The feasibility and effeciency of the algorithms mentioned above are all described and discussed with simulation results in detail.Based on above framework and fusion algorithms, conceptual model of solution knowledge space is described with knowledge space theory. The evolution of knowledge nodes and local pattern in solution knowledge space is discussed to adapt its global structure. The clustering and searching strategies in that space are also analyzed to improve efficiency of the whole fusion process.Actually value of new knowledge objects should be judged by applications, but there is not an effective evaluation method in existing knowledge fusion systems. Inorder to overcome it, a feedback-adjust and self-adapting mechanism is provided in the dissertation. A set of parameters whose evolution process is described in formal way are proposed to reflect the evaluation results of knowledge objects. The results of simulation test show its rationality.After theoretical discussions, two applications based on above framework and algorithms are displayed. The case studies in "Networked manufacturing-oriented collaborative customization service" and "Automatic question-answer service in communication domain" show how to integrate our knowledge fusion framework into applications.