Abstract
Because of the spread of microblogs and SNS (Social Networking Service), life logs submitted by their users are becoming various. With analyzing these life logs, attributes of those users such as gender, age, occupation, and living region can be acquired. So it is being expected that these attributes can be utilized in corporate sales strategies, etc. In particular, it can be said that the living region is very useful in area marketing. In existing researches for estimating living regions, methods using the region’s special words selected from topic models or analyzing differences in preference for each user's living region have been proposed. However, in the first method, the appearance frequency of topics concerned with a season such as snow is different from another season. Therefore, here comes a problem that the appearance frequency of region’s special words extracted by this method in different seasons are different. And then the region which special words stands for is different from living region. On the other hand, the second method also has a problem that it is difficult to estimate living region because of the low correlation between it and preference of a user. Therefore, in this research, we consider a method focused on synonyms which are used in different regions have the same meaning.
