In recent years, many kinds of geographic information in socio-demographic attributes such as land use or population is available for fine aggregation unit. In this research, we pointed out the similarity between factor analysis and topic model, and applied both models to geographical information with multiple sociodemographic attributes in fine unit, in order to clarify the advantages of the latter. The 12 attributes of sociodemographic data was collected at the tertiary mesh level from the five capital cities of the prefectures in Chugoku area. In order to process the raw data into the input for topic model, the dataset with 12 attributes were discretized into totally 96 class attribute by natural class classification. As a result, eight geographical topics were obtained from the topic model, but only three factors were obtained from the factor analysis. The spatial distribution of dominant geographical topic which is only available from the topic model can appropriately capture the land-use characteristics of the actual cities.
View full abstract