No abstract

Abstractive text summarization is one of the most interesting problems in the research field of Natural Language Processing. Recent advances in sequence to sequence model have made it possible to apply new approaches for abstractive text summarization and perform significantly. But most of the existing systems suffer from some drawbacks like word repetition, producing inaccurate or irrelevant information etc. In this work we propose a novel architecture incorporating advanced word embedding layer and topical feature with a pointer generator network to generate more topic oriented summaries in a logically sequenced way. Adding a word embedding layer with the model can capture semantic features of words in the input sequence more accurately. Also our proposed system with incorporated topical features ensures that the summaries focus on the most important parts of the source document. We applied our model to the CNN/Daily Mail dataset and outperformed the baseline model by all the ROUGE scores. DUJASE Vol. 6 (2) 39-48, 2021 (July)