Research On Measure And Evaluation Of Decision Implications And Inference Rules

Knowledge representation and mining is an essential research in artificial intelligence,with a wide range of application prospects.As a form of knowledge representation,decision implication is simple and clear,and thus has strong reasoning ability and a good performance in solving decision-making problems in artificial intelligence.In recent years,decision implication has been paid more and more attention and used by researchers.At present,the research on decision implication is based on no noisy data.Since there is no completely denoised data in natural life,it is likely to be affected by noise when using decision implication for knowledge representation and inference in real scenes,resulting in poor robustness and low use value.To solve this problem,it is necessary to introduce measures that can be used for knowledge representation and inference of decision implication and to filter out the decision implications with better robustness and higher value.In this paper,we conduct studies on the measures and evaluation of decision implications and inference rules to obtain decision implication with good robustness and high value.The details of the research are as follows.(1)Studying robust decision implication based on the support measure.The support measure is introduced to decision implication,and by analyzing the changes of support values before and after using inference rule,we measure and evaluate inference rules.Furthermore,we introduce the decision implication canonical basis that satisfies the support threshold,and propose the method of obtaining the canonical basis.Finally,the method of determining the support threshold is also given and verified by experiments.In addition,we propose a method to obtain the decision implication closed set satisfying the support threshold by reasoning with the canonical basis,and verify the superiority of this method on different data sets,and the robustness of the decision implication obtained.(2)Studying robust decision implication based on the support and imbalance ratio measures.Since empty objects may affect the support measure,we introduce into decision implications the imbalance ratio that satisfies the invariance of empty transactions.By analyzing the relationship between the two measures,we determine the range of the imbalance ratio threshold.On this basis,the imbalance ratio is also used for measuring and evaluating inference rules.Furthermore,we study the canonical basis satisfying the thresholds of two measures.The relationship between the two canonical bases is clarified,and the method of generating the canonical basis is proposed.In addition,the factors affecting the time efficiency of the method are also studied in depth,and an optimization method,called DLSI+ algorithm,is proposed.