Study And Implementation On The Algorithm Of Mining Cognitive Maps Based On Data Resources

Cognitive maps, an effective tool of knowledge management and expression, can describe the entities in reality and their causal relationship intuitively in the visualized form of graphs. It has caught more and more attention in the academia and has gradually become a new direction in the artificial intelligence research thanks to its visualized knowledge expression ability and powerful matrix-based reasoning mechanism.Traditional methods of establishing cognitive maps, which rely largely on expert experiences, have two defects. On one hand, they ignore the useful information stored in the accumulated data resources in the related fields; on the other hand, the cognitive maps based on the expert experience, which inevitably has certain defects such as subjectivity and one-sidedness, accordingly are bound to deviate from the reality.This thesis, based on data resources, proposes new methods of establishing cognitive maps by training data samples and thoroughly explores the three steps in the whole process of mining—preprocessing data, selecting nodes and mining. As to the first step, this thesis proposes a new algorithm to detect noise data, which is efficient compared with traditional methods. As to the second steps, it proposes a new algorithm based on rough set theory. According to the experimental results, this algorithm can eliminate the attributions that fail to describe the key features of the reality and get the exact set of attributions. As to the third step, this thesis, based on the qualitative and quantitative analysis of the causality between the nodes, proposes a new algorithm to mine for cognitive maps more informative than those established by the traditional methods.At last, this thesis designs a cognitive map mining system on the basis of data resources, and then demonstrates the function modules and operation process through an actual case. The result of the case shows that this system fully taps data resources when mining for cognitive maps, and that, without reference to expert knowledge, it can establish cognitive maps which accurately express knowledge stored in data resources.