Automatic Knowledge Graph Construction over Efficient Information Extraction Networks

Knowledge graph construction(KGC) aims at constructing a knowledge graph from enormous amounts of unstructured sources. Once unstructured information is stored in a graph-structured format, it becomes easier to access and utilize. In this paper, we propose a automatic knowledge graph construction method based on an efficient contextual model, which is composed of layered convolutional neural networks(CNN) and attention-based bidirectional LSTM layers. The model achieves similar performance compared to the RoBERTa(pretrained model) but is time conserving and is lighter than this kind of pre-trained model. In addition, we propose a framework to enrich the constructed knowledge graph(KG) with external high-quality sources using relation linking. The refined KG is visualized based on Neo4j. Our experiments highlighted that our method achieved a few points of amelioration that can be readily applied to other similar domain fields. Knowledge graph database holds the potential to contribute to the advancement of educational initiatives in the field of music e ducation, thereby enhancing educational prospects.