Exploring Multi-Meta Information Embedding for Recognizing Complex Named Entities in a Mechanics Intelligent Tutoring System

Named entity recognition (NER) is a bottleneck in the development of intelligent tutoring system (ITS). NER is difficult in ITS due to the complexity and the lack of entity boundary of concepts that must be named. Compared with the open domain, ITS domains have a large number of specialized vocabulary items and long entities, which leads to a low recognition performance. In this study, we propose a collaborative model for ITS NER in the mechanics domain using Chinese texts. First, a multi-meta information embedding (MMIE) method including character embedding, lexicon embedding and radical embedding is proposed to capture the semantic information of characters for Chinese sentences. Second, a collaborative graph neural network (GNN) framework is presented to fully integrate the multi-meta information and learn the contextual information of each character. Then a collaborative transformer structure is employed to recognize the named entity. Finally, extensive experiments are carried out, and the experimental results indicated the proposed method outperforms previous methods in our collected dataset and three mainstream datasets. In practice, a series of ablation experiments and case studies implied that our method is effective and efficient for Chinese mechanics NER.