A Knowledge-Embedding-Based Approach for Process-Decision Knowledge Mining

Abstract

Association rule mining algorithm is a main technique to discover and extract knowledge of rules represented with antecedents and consequents, which can be applied to mining process-decision knowledge. However, it is difficult to reflect the semantic association between process data items when only the support degree and confidence coefficient are used as indicators for judging whether the data items can be included in the content of knowledge. This usually results in amounts of invalid rules in output. To address this problem, a knowledge-embedding-based approach is presented in this study. To integrate semantic information into algorithm, a matrix of semantic correlation among data items is developed as the knowledge to be embed. The indicators of cohesion and relevancy are introduced based on the matrix of semantic correlation. Then two fusion models are developed to enhance the semantic correlation evaluation, one is built with the cohesion and support degree coupling while the other is proposed with the combination of relevancy and confidence coefficient. Based on these models, the proposed algorithm is able to identify the effectiveness of the composed knowledge expressed in rule form effectively. This method is carried out to mine process-decision knowledge with the machining data of gears and shafts. The results reveal that the number and proportion of correct rules are improved with knowledge imbedded, which validate the proposed method.