A Collaborative Learning Grouping Strategy with Early Warning Function Based on Complementarity Degree

Zhizhuang Li, Zhengzhou Zhu

Organizing groups is a critical process in implementing cooperative learning. The grouping strategy based on the degree of complementarity is a popular grouping strategy at present. However, the existing collaborative learning grouping strategy based on the degree of complementarity has disadvantages such as insufficient modeling accuracy for students‘ ability and lack of rationality for the reasons of regrouping. This paper proposes a collaborative learning grouping strategy with early warning function based on the degree of complementary mastery of knowledge points. First, we take knowledge points as the minimum unit, and use linear regression and expectation maximization algorithm to accurately model each student’s mastery of each knowledge point. Then we use the inverse clustering algorithm based on knowledge points to classify students. Finally, we use LSTM neural network to predict the scores of each group in the next week, and early warning was given to the groups with significantly reduced predicted scores, and targeted suggestions were put forward for them according to the types of the warned groups. Experimental results show that the grouping strategy proposed in this paper can effectively improve the learning effect of students. The average precision and average recall of LSTM based group early warning were 30.1% and 27.6% higher than that based on linear regression, respectively.