Social Media Event Summarization using Neural Networks

Abstract

Micro blogging sites have become extremely popular, with a huge number of tweets being posted daily on a broad assortment of subjects. Recent studies have exposed that a significant part of these tweets are regarding "events", and the discovery of new events in the tweet-stream has pulled in a lot of attention. In the Natural Language Processing (NLP) field, automatic text summarization is viewed as an extremely tricky problem. The content substance on the internet, specifically, is rising at an outstanding level. The capacity to decode through such immense volumes of data, to facilitate mine the require knowledge, is a foremost undertaking and needs an automatic system to serve with the present warehouse of data. A decent text summarization framework should comprehend the entire content, reorganize information, and create coherent, useful and amazingly tiny summaries to pass on the significant information of the innovative content. In this article, an inventive content summarization framework has been built, which combines RNN with LSTM and encoder-decoder structure. Our framework is assessed on the LCSTS dataset, which is an excellent corpus of Chinese short content summarization dataset created from "SinaWeibo". Empirical outcomes exhibit the adequacy of the proposed framework, which outperforms all the existing approaches. Particularly, the proposed framework enhances the semantic consistency by 5% in terms of human assessment.