A Joint Course Knowledge Entity and Relation Extraction Method for Educational Data

摘要

Considering the current growth of data in Educational Field, especially in recent times, due to the COVID-19 epidemic, the volume of online course data has increased exponentially. How to extract useful entities and relations from mass educational data has turn into an academic hotspot. In traditional models, entity recognition and relation extraction are considered as two separate subtasks, which may cause error propagation. Therefore, the development of a more effective and accurate entity and relation extraction model in Educational Field is being widely studied. Here, we propose an entity and relation joint extraction method based on deep learning in Educational Field. The proposed model (XMMC) includes XLNet, a pretrained language model for obtaining text word vectors, Mogifier BiGRU neural network for obtaining text context information, Multi-headed Attention for focusing document key information, and CRF for extracting text entities. The obtained entity context information and entity label are further passed to the sigmoid function, which can enhance the accuracy of the subsequent relation extraction and achieve the goal of joint processing. The proposed model performs better than the existing model in terms of accuracy rates (P), recall rates (R), and F1 values (F1 scores).