In this paper, we combine the interactive feature capture and basic feature capture modules of multilayer perceptron to extract and fuse features of knowledge concepts. With the help of cluster search algorithm and softmax way to optimize the learning path, the learning path recommendation model based on multilayer perceptron is proposed. Suitable datasets and research objects are selected, and the effectiveness of the blended teaching mode supported by the TMLPE model is examined through experiments. The TMLPE model scores higher than the baseline model in MRR and NDCG indexes, and the Er index is 12.23% higher than that of the ELPRKG model with the second best performance, which is capable of generating accurate and diversified learning paths for learners. The different models of teaching activities show significant differences at the 0.01 level (p=0.003.42, 4.06>3.15, and 3.96>3.11. The pvalues of the personalized learning needs and the personalized learning goals as a whole are all less than 0.05 and the R-values are all in the interval of 0.5 to 1.0. The intelligent recommender system based on the multilayer perceptual machine model in this paper can effectively meet the learning needs of different students, and has a positive role in promoting the development of blended teaching for English majors.
