Abstract: [Purpose/Significance] Mining college student data and constructing studnet profiles is conducive to in-depth understanding of students' needs, improving management level, and promoting intelligent service. [Method/Process] Based on the multi-source data mainly generated by the management and service process of colleges and universities, student profiles were developed by focusing on consumption, academic and social indicators, analyzing the characteristics of students, using the Scikit-Learn tool of Python, and applying the K-means clustering algorithms. Empirical research was carried out and representativeness of student portraits from individual and group perspectives was studied. [Results/Conclusions] First, this paper attempts to utilize a new data fusion perspective, by fusing explicit data with implicit data, and generating three-dimensional indicators of consumption behavior, academic behavior, and social behavior. Secondly, in order to solve the problem of single application scenario in previous research, the method of user profile construction is used to realize the fusion of multiple scenarios. Finally, based on the real student data, this study uses K-means clustering algorithm to select groups of students with different characteristics on the basis of previous research. The data of college students is analyzed, and further empirical research is carried out to describe the "consumption-academic-social" profiles of college students. Constructing student profiles from the perspective of multi-source data fusion can effectively provide a basis for decision-making by different units in colleges and universities, such as academic affairs,. Especially in the post-epidemic era, the profiles of college students can detect potential risks in time. The study found that at the individual level, by interpreting the label information of students' portraits, it is possible to understand the 3 aspects of students' consumption, academics and social interaction, and realize dynamic monitoring of individual students. At the group level, through cluster analysis, students with different characteristics can be selected, especially in terms of consumption behavior, and the characteristics of students' activity and stability can be deeply analyzed, which can not only provide a basis for the macro-level observation of students, but also provide new ideas for exploring the correlation between different behavioral elements of students. At the application level, the integration of multi-scenario student profiles can simultaneously realize abnormal identification and early warning, group attention and guidance, and resource planning and adjustment, which greatly broadens the application scenarios of research and improves the energy efficiency of education and teaching management in colleges and universities. However, due to the limitations of data and algorithms, the accuracy and ease of use of student portraits still need to be improved. There are both constraints from practical conditions and insufficient research methods. In future research, more extensive research should be used to improve college student profile construction system, and constantly develop more suitable techniques.

Key words: student profile, consumption analysis, social analysis, academic analysis, K-means clustering, information behavior