Construction of digital art knowledge graph based on deep recurrent neural network

This study presents a method for constructing digital art knowledge graphs based on deep recurrent neural network (DRNN).A digital art knowledge graph is initially constructed by extracting visual features with ResNet50 and identifying textual entities via a CNN-BiLSTM-CRF model.Then, a DRDA model with bidirectional gated recurrent unit (GRU) and neighbour-aware attention is proposed for graph completion.Experiments on DBPedia50k and DBPedia500k show DRDA's superiority over three baselines.On DBPedia50k, DRDA improves head prediction MRR by up to 55% and achieves the lowest MR in tail prediction, though trailing slightly in Hits@10.On DBPedia500k, DRDA consistently outperforms baselines with MR reductions of 59-406 and MRR gains of 2%-19%.Further analysis identifies optimal depth and neighbour parameters, validating the model's scalability and its effectiveness in capturing complex semantic dependencies in large-scale multimodal art data.