Deep Learning-based Erroneous Data Detection in Linked Data Context Using LSTM and Embeddings

RDF (Resource Description Framework) triples are the fundamental building blocks of knowledge graphs and semantic web applications. However, incorrect triples can lead to inaccurate and unreliable data, which can adversely impact applications. This study proposes a deep learning model to effectively detect the correctness of RDF triples. The proposed model learns the patterns and dependencies within RDF triples by embedding the subject, object, and relation into dense vectors and capturing sequential and long-term dependencies using LSTM (Long Short Term Memory) layers. The outputs of the LSTM layers are concatenated and fed into a dense layer with ReLU (Rectified Linear Unit) activation for non-linear transformations and feature extraction. Finally, a sigmoid-activated output layer performs binary classification to predict the triple’s correctness. Experimental results demonstrate the effectiveness of the proposed model, achieving a high accuracy on the validation set. This improved data quality and reliability can benefit applications reliant on RDF data. The proposed model provides a novel and effective approach to RDF triple correctness detection. Its superior performance suggests that it can be generalized to other RDF datasets.