The main goal of Starry system is to analyze and formulate student labels based on the multi-dimensional data of students collected by various intelligent terminals and perceptible devices in the smart campus, generate visual portraits of students' attributes in various dimensions, and predict students' behavior trends, so as to make early warning of abnormal situations. The system can timely and accurately understand the situation of students, and timely make early warning of problems, to help students pay attention to their own living conditions, find themselves in life is not easy to detect changes, so as to make evaluation and improvement. The Starry system not only paints a unique portrait of each user, but also integrates multiple individuals to obtain group behavioral trends. The subject of this behavior can be a user or a group of people with the same characteristics. The Starry system can more intuitively reflect the most essential changes in the data subject, and all parties can visually see the specificity of the task through the generated visual portrait. Students can get the student's learning and living status according to the various dimensions of their school behavior data, and help themselves find the reasons for the decline in academic performance; teachers can perceive the changes in all aspects of students in the behavior trend of students, so as to provide accurate help for students; managers can obtain the trend of students' behavior from students' portraits, so as to provide better management and service for students' learning and life. The system selection is based on the Hadoop processing framework to build an experimental environment. Since student personal data comes from various business systems, ETL needs to be used to synchronize data from different heterogeneous data sources. This experimental platform is built on Aliyun ECS server. First, hadoop pseudo-distributed cluster environment is installed, and ETL data acquisition and processing tools are used to synchronize and integrate data from different heterogeneous data sources into Hive data warehouse, and data files are stored in HDFS distributed system. Then use Hive data warehouse to store data, and use SQL-like statements to simply clean and preprocess data. At the same time, the parallelization of Canopy kmeans algorithm on Spark is realized by scala, and the technology is applied to the distributed processing of historical data to complete the clustering segmentation of student behavior and improve the clustering efficiency. Finally, Java Web technology and ECharts are used to visualize the analysis results and provide user-friendly query interaction. The college student portrait system based on Hadoop applies the big data user portrait technology to the comprehensive quality evaluation of students, provides a strong technical support and management mode for the method, process and feedback of student evaluation, and ensures the authenticity and systematicness of the comprehensive quality evaluation factors. In the application of student portraits, the system has added the functions of early warning of students' abnormal behavior and judging groups in schools with financial difficulties to help university administrators understand the situation of students more timely and accurately. Through data analysis and integration, every student can get accurate and personalized guidance in the Starry system, adding more possibilities to university life; every teacher can make targeted teaching according to the students' personality, explore and cultivate students' specialties and improve the efficiency and quality of teaching, so as to provide more possibilities for students' growth. Starry system can provide accurate and personalized guidance for students, provide a new way for the scientific management of students, and provide new ideas for the intelligent management of campus.