On the basis of quantifying the learning process and integrating learning data, learner feature analysis and extraction, multi-dimensional learning state recognition of learners, and construction of personalized learner portraits are used to achieve steady-state modeling of learners based on multi-source data. Learner modeling is a typical dynamic stochastic process. In order to construct a dynamic model that covers key learning features and their spatiotemporal trends, a research route of "learner feature analysis and extraction multidimensional learning state recognition personalized learner portrait dynamic construction learner collective portrait and evolutionary analysis" is adopted.