HTS-DL: Hybrid Text Summarization System using Deep Learning

Abstractive text summarization is the task of creating a summary from a document by merging facts from different sources and make a short description of them. In this procedure, the meaning and the content information should be kept. In this paper, a hybrid summarization system using deep recurrent neural network is proposed, which can create new sentences by information extracted from the text The proposed model is the combination of extractive and abstractive summarization and has the encoder-decoder structure. The encoder extracts information from the source document and encodes this information in a compressed representation. The decoder takes the encoder’s output as input and generates a summary, which has an acceptable semantic and syntactic structure. Experimental results show that the proposed model could improve both the performance of abstractive summarization and the time of training. This model does the single-document multi-sentence summarization and does not have any dependency on language. Therefore, it can be used for other languages without any modification in future.