The red line in Fig. 9 is the time curve of normalized entropy values for all users, and other lines are the normalized entropy values of every interest clusters for every hour. The average value of normalized entropy of all users is 0.74, which is bigger than any interest cluster. This implies that users in the same interest cluster tend to concentrate on limit application categories and our cluster method works. In addition, different interest clusters with distinct application category preference have very different time curve. Namely, application usage behavior between distinct interest clusters is quite different. In order to see the normalized entropy values for each interest cluster more clearly, we further draw three graphs to show the difference between different interest clusters in Fig. 10, Fig. 11 and Fig. 12.