Exercise recommendation is an important research direction in intelligent education. The most advanced exercise recommendation methods focus on the association between each student and each knowledge concept and exercise, and model the exercise recommendation as a Multi-Objective Optimization Problem (MOP). However, this method is based on each individual student to recommend exercises, which will cause the recommendation process to consume too much time when the number of students is large. One approach to solve this problem is to recommend exercises based on student clusters after clustering students with similar characteristics. According to the specific characteristics of the student group, this paper proposes an Adaptive Hierarchical Clustering algorithm with Ward’s method. The algorithm can adaptively adjust the within-cluster variance threshold for different student sets to make appropriate clustering for them. After clustering, the exercise recommendation process, which is modeled as a MOP, will be carried out in clusters. The experimental results show that when using different Multi-Objective Evolutionary Algorithms (MOEAs) to solve the exercise recommendation process, the recommendation performance after clustering is improved compared with that before clustering, and the total recommendation time for the student group is greatly reduced.