Abstract: In order to make effective use of log files and depict learners' portraits more deeply, a two-way clustering modeling method(TWC) is proposed and used to analyze and mine the data of learning behavior of more than ten thousand learners in the College of Network Education of a university. Considering the implicit characteristic of educational data, this method takes fine-grained data as the core, obtains the categories of learners in different models through two-way clustering calculation, and finally describes learners based on the fused model. The experimental results of four classical clustering algorithms and the TWC algorithm and the clustering results of the TWC algorithm show that the TWC algorithm can enhance the cohesion of clusters and cluster learners more accurately and describe the learners' profiles more deeply and comprehensively.