Content curation algorithm on blog posts using hybrid computing

Content curation is a significant step to identify the relevant content for the searched topics. There are many methods introduced to generate summarized contents but those methods focussed only on generating precise contents that lacked the key essence of the input texts. Therefore, we propose a hybrid model with the integration of self-attention to the bi-directional long short-term memory auto-encoder (Bi-LSTM-AE) to generate information-rich abstracts. Initially, the dataset is pre-processed and then the major word-level and sentence-level features are extracted. Then, based on the similarities between the contents, the extractive summary is generated which is then given to the auto-encoder for final abstraction. The efficiency of the model has been proved through simulations with the CNN/Daily Mail dataset in terms of ROUGE metrics. The proposed model outperformed the other compared models with a score of 0.59 for ROUGE 1, 0.39 for ROUGE 2 and 0.71 for ROUGE L with high generalization.