The conversion from similarity for mobile life-log to euclidean distance

Mobile devices are widely used due to their convenience. However, information from Mobile devices always contains the semantic contents, which are hard to be analyzed by the Data Mining methods such as clustering. Therefore, some researchers calculate the relative similarity based on k-means clustering or association rules. However, without the consideration of features of non-Euclidean data, and the impossibility of converting the relative similarity to Euclidean space, the accuracy is low. In this paper, we calculate the similarity and Term Frequency Inverse Document Frequency (TF-IDF) based on analysis of mobile life log data, and then convert the similarity with the non-Euclidean features into the distance on the Euclidean space by pseudo-Euclidean embedding. As a result, we realize the data mining for non-Euclidean data. We apply the proposed method in collecting data that reflect the life-patterns of students from mobile devices. Experimental comparisons with real life-patterns of students show the feasibility and effectiveness of the proposed method.