Ext-ICAS: A Novel Self-Normalized Extractive Intra Cosine Attention Similarity Summarization

With the continuous growth of online news articles, there arises necessity for an efficient abstractive summarization technique problem information overloading. Abstractive is highly complex and requires a deeper understanding proper reasoning to come up with its own summary outline. task framed as seq2seq modeling. Existing methods perform better on short sequences; however, long sequences, performance degrades due high computation hence two-phase self-normalized deep neural document model consisting improvised extractive cosine normalization phases has been proposed in this paper. The novelty parallelize sequence training by incorporating feed-forward, network Extractive phase using Intra Cosine Attention Similarity (Ext-ICAS) sentence dependency position. Also, it does not require any explicitly. Our Bidirectional Long Short Term Memory (Bi-LSTM) encoder performs than Gated Recurrent Unit (Bi-GRU) minimum loss fast convergence. was evaluated Cable News Network (CNN)/Daily Mail dataset average rouge score 0.435 achieved also computational reduced 59% number similarity computations.