Abstractive Text Summarization For Multimodal Data

Abstract

The availability of data from different sources such as newspapers, images, online articles, social media, etc is rapidly increasing. As a result, finding the most relevant article according to one's requirement has become a time consuming and tedious task because it is not possible to read each article and then decide which particular information would be the most useful. This problem can be solved with the help of an abstractive summarizer that would generate concise, easy to understand summaries of those articles which would be an approximate representation of human written summaries. The proposed system generates abstractive text summaries of contents existing in multi-modal data formats, namely, text files, image files and video files using the Recurrent Neural Network model with encoder-decoder architecture and attention mechanism. The paper also discusses various text extraction methods, the implementation of the mentioned RNN model on a particular corpus and its results.