Multi Modal English Vocabulary Inference Algorithm Integrating ELECTRA and Knowledge Graph

In current English learning, the improvement of vocabulary inference ability is crucial to learners” language application ability. However, the existing single-mode algorithms are often unable to fully tap the deep relationship between words, resulting in insufficient accuracy of inference. This article aims to propose a multimodal English vocabulary inference algorithm that integrates ELECTRA model and knowledge graph to improve the accuracy and efficiency of vocabulary inference. Firstly, the ELECTRA model is used for deep learning of contextual semantics, and then a knowledge graph based vocabulary relationship network is constructed to integrate the semantic and relational information of vocabulary. Next, by integrating the contextual representation of ELECTRA with the node features of the knowledge graph through multimodal feature fusion technology, an inference model is constructed for training and optimization. The experimental results show that the algorithm achieves an accuracy of 87.5% in vocabulary inference tasks, which is more than 8% higher than traditional methods, and also improves inference efficiency by 25%. The multimodal inference algorithm that integrates ELECTRA and knowledge graph effectively solves the problem of information silos in vocabulary inference, providing English learners with more accurate and efficient vocabulary learning tools.