Unstructured data processing is a frequent problem with text processing systems. Some of the works are generated RDF statements from unstructured textual data. Extracting the correct feature from the textual data includes some issues with existing works. Efficient feature extraction approach and RDF statement generation technique are necessary to address these issues. This paper introduced an effective hybrid deep learning model, generating RDF statements from a given unstructured textual data. This paper uses data from publicly available datasets namely BBC News and Lonely Planet dataset. The datasets are pre-processed using some approaches such as data normalization, Case folding, Stop word removal, sentence segmentation, and PoS (Parts of Speech) tagging. The pre-processed data is fed into the feature extraction process whereas the corresponding entity and attribute feature values are extracted using the Assimilated N-gram method. From the extracted features, the optimal features are selected using the Walrus optimization (WaOA) algorithm. The Triples are classified using Soft Attention based hybrid ResNet-Bidirectional Long Short Term Memory model (SAtRes_BiLSTM). Finally, the attained triples are then converted and saved in RDF format. The performance of the suggested RDF statement generation is related to diverse emerging methods to prove the efficiency of triple extraction. The proposed method achieved 95%, 91.2%, 90%, and 93.2% respectively in terms of Accuracy, Precision, Recall, and F-Measure.