This study proposes a method for constructing and analyzing the knowledge graph of English learners in colleges and universities based on the local linear embedding LLE algorithm, which optimizes personalized learning support through dynamic characterization and cross-domain knowledge migration. The dynamic knowledge graph covering 9673 nodes is constructed with the core literacy of English discipline as the orientation, and the adjacency matrix (dimension 9673×9673) and 512-dimensional feature vector are generated using the feedback data of exercises. The joint knowledge migration method HLLEJKT is proposed to achieve cross-linguistic representation space alignment through kernelization extension and label optimization, and achieves 75.13% and 85.75% accuracy in XNLI and STS 2020 benchmark tasks, respectively, which is an improvement of 55.69 and 124 percentage points compared with the traditional method mBERT. In practical applications, the graph-based intelligent retrieval system achieves an average accuracy rate of 96.55% in the retrieval of teaching resources in English, Chinese, Spanish, and Arabic, with an accuracy rate of 98.47% in English and 94.56% in Arabic, and the length of the retrieval path is shortened to 21.60, which is 17.40% lower than that of the traditional method of fuzzy retrieval path for teaching resources. The method effectively integrates semantic association mining and knowledge migration mechanism, providing a theoretical breakthrough and practical paradigm for multilingual education technology.
