This article proposes an automatic generation method of a relation matrix for Interpretive Structural Modelling (ISM). In ISM, keywords related to the problem are selected. Then the relation between two arbitrary keyword pairs is examined. When there is a relation between these two keywords, a non-zero value such as 1 is assigned to this relation. After examining the relations of all pairs of keywords, an n×n relation matrix is generated where n stands for the number of keywords. Then a reachable matrix is calculated from the relation matrix. By using the reachable matrix, the hierarchy structure of the problem is constructed. Building a relation matrix requires much time and labour. This article proposes an automatic generation method of relation matrix based on the co-occurrence probability of keywords in primary materials. As the proposed method is fully automatic, less labour is necessary than in the original ISM method. This method is evaluated by using newspaper articles.