Hierarchical Attention Multi Model Deep Learning for Multi Document Summarization

Nowadays, with the progress in the different fields of science and technology, the massive number of information is easily available in the internet. It helps users to gather abundant information quickly based on their requirements however this caused severe barriers in reading, and difficulties in accessing usable information. Thus, different summarization techniques are generally followed for the retrieval of relevant information from the source documents. Multi-document summarization helps to summarize text from multiple set of documents, but the techniques used for text summarization are limited and also suffers from text redundancy issues during summarization. Hence, a Blue Monkey Integrated Coot Optimization-Hierarchical Attention Multimodal Deep Learning (BMICO-HAMDL) model is designed in this research for the summarization of multi-documents. The tokens from the input multi-documents are extracted using Bidirectional Encoder Representations from Transformers (BERT) tokenization and different feature extractors are used to extract the features from the tokens. The extracted features are subjected to Multi-document summarization using HAMDL, where the BMICO is used to fine-tune the optimal weights of HAMDL model. In addition, the experimental results show that the BMICO-HAMDL achieved superior performance with maximum rouge, recall, precision, and F-measure of 85.88%, 96%, 94.79%, and 95.39% respectively.