Convolutional Block Attention Assisted Dense Stacked Bi-LSTM for the Generation of RDF Statements

The rapid enhancement of the World Wide Web has attained an immense evaluation of information sources over the Internet. Due to heterogeneity and improper structure of web information sources, accessing significant information has resulted in redundant data. Hence, it is highly necessary to process the unstructured textual data for the better generation of the Resource Description Framework (RDF). Initially, the collected text data was pre-processed using one-hot encoding, z-score normalization, tokenization, stop word removal, and POS tagging. From the pre-processed data, effective features are extracted using the BERT-LSTM model (BLM), whereas similar features can be clustered using the Adaptive density K-means clustering (ADK-MC) mechanism to minimize the model complexity. The RDF triples can be classified using the convolutional block attention-assisted dense stacked Bi-LSTM (CBD-SBiL) model. The proposed method achieved better performance with 98.09% accuracy, 99.06% precision, 98.36% recall, and 98.51% F-measure.