In the previous paper., the correlational analogy method, in which we find an associational relation between two assignmemts in plan, was proposed in order to assist design process. In this paper, we discuss statistical fundamentals on the finding method of associational relationships. Furthermore, we propose a method keeping the consistency of the architectural knowledge-base. The associational relationship (A, B) means that the two assignments A and B occur coincidentally in many plans. Then, we can use the statistical testing method as a criterion of judgement whether the relationship is meaningful or not. That is, we can find only meaningful associational relationship between two assignments using the fact that a random variable defined by correlation coefficient between two assignments has a t-distribution. Although associational relationships passed the statistical test, it is not always to keep the consistency in the logical relations among relationships. For example, when relatinships (A, B), (B, not C) and (A, C) have been passed, the first two relationships infer a relationship (A, not C) which contradicts the third relationship (A, C). We discussed this problem using the graph theory, and find a fact that if the graph expressing the associational relationships is "a balanced graph" then there is no contradiction among the associational relationships. Then we propose a method keeping the consistency by using this fact.
View full abstract