In this paper, based on the characteristics of English vocabulary and students’ behavioral data, a cognitive level test learning model based on IRT theory is constructed, and for the defects existing in K-means clustering algorithms, a user clustering recommendation algorithm based on the minimum variance is obtained by using the minimum variance optimization initial cluster heart method. On this basis, a personalized recommendation platform for English vocabulary learning based on students’ vocabulary level with dichotomous K-means clustering is designed and implemented, and the effectiveness of the platform is verified. The experimental results show that the method proposed in this paper can very directly observe that the topic parameters and students’ learning ability values match the information reflected in the actual data. In addition, the accuracy of this paper’s model in successfully recommending learning resources can be improved by up to 58.23% compared with the traditional model, and relevant extended knowledge of non-vocabulary subjects such as oral expressions is given in the recommendation results, which alleviates the problem of increasingly narrow vision of students caused by the cocoon effect. Teaching experiments show that this strategy can significantly improve students’ learning of English vocabulary and increase the mean value of their English scores by 7.4628 points. Obviously, the model in this paper solves the defects of the existing English vocabulary learning software that does not meet students’ individual needs.
