Based on the DBSCAN analysis, kernel density analysis and word cloud analysis, this paper analyzed the composition types and spatio-temporal evolution characteristics of catering points and catering clusters form macro, meso and micro dimensions within 5th ring road in Beijing with the caterings POI data in 2012, 2014 and 2016. The conclusions are as follows: first: (1) The number of total catering POI increased by about 6 times, and the center of the whole data shows a tendency to move from west to southeast. There is little change of quantity of total catering clusters. The distribution is relatively uniform. The scale of single catering cluster grows significantly, and the large-scale clusters have contiguous trends and the number of clusters in Chaoyang has been the largest, in Dongcheng and Xicheng varies at least. (2) The large-scale catering clusters are mainly distributed in the CBD, Sanlitun, Nanluoguxiang, Wangfujing, Zhongguangcun, and the scale of some clusters fluctuates. As the proportion of composition of main clusters, the Chinese food and casual dining continues to increase, while the western food decreased, there are similar trends in the type of catering. (3) The diversity of catering brands has increased year by year, among them, the proportion of small dining sites such as Shaxian snacks has increased obviously, and the proportion of fast food such as KFC has declined slightly but still considerable, and the time-honored brand such as Quanjude accounted for a downward trend. Second, the results obtained by POI data of time series are more objective and accurate. It is a meaningful attempt of data mining and analysis in the era of big data.