| 摘要 |
PURPOSE:To save a data storage area of a computer by performing such a clustering operation where the past clustering results and their subsequent statistic values are stored and then used for each addition of plural new data and therefore the total description length is minimized to all data. CONSTITUTION:The data described in plural attributes are inputted one by one and a probability parameter subsequent to the clustering is calculated to each clustering that can be attained with division or integration of the past clustering operations. Then the total description length necessary for description of the data and class dividing structure is calculated based on the value of the probability parameter. The total description lengths of each clustering are compared with each other, and the clustering of the minimum total description length is outputted as a new clustering. Each of these steps is carried out with the sequential fetching of data, and the clustering form is outputted for each input of data. Thus a data storage area can be saved in a computer. |
