Research on Knowledge Acquirement Methods based on Kolmogorov Complexity

1.Kolmogorov Complexity is applied to the optimization of rules in theknowledge acquirement domain. The objective function for optimization isdetermined based on the Minimum Description Length principle and isattained by algorithms of rules extension and local search.2.Take part in the design and construction of our Scientific literaturebased Protein Interaction Extraction System (SPIES), which extractsprotein-protein from bio-medical literatures by the modules of pre-procession,rules auto generation and optimization and dynamically matching. Theimplementation of our rule optimization algorithm improves the systemperformance by 8.9%, while greatly reduces the number of rules.3.Define the conditional information distance, and infer its formula fromthe normalized information distances. Through experiment of a multi-senseword, it is proved that conditional information distance can describe therelationships among conceptions more clearly, accurately and flexibly thanoriginal information distance.4.The approximate calculation of Kolmogorov complexity is proposedbase on the coding theory. Then, NSD (Normalize Statistical Distance) andCNSD (Conditional Normalized Statistical Distance) are calculated. NSD isproved to have better ability in processing concepts than Compressed-basedDistance Measure (CDM).5.The extended CNSD is proposed for overcoming the indifinity of thecondition in CNSD. First the semantic relationship for detection is determined,then, it is converted to patterns. Finally, CNSD is calculated from thesepatterns. A conception classification system and an intelligent questionanswer system are constructed to prove the power of extended CNSD.