Meeting Summarization, A Challenge for Deep Learning
Résumé
Text summarization is one of the challenges of Natural Language Processing. Given the volume of texts produced daily on the Internet, managers can no longer have an exhaustive reading of current events, or progress reports from their employees, etc. They urgently need tools to automatically produce a summary of this flow of information. As a first approach, extractive summarization tools have been produced and there are now commercial tools available. However, this family of systems is not well suited to certain types of texts such as written transcriptions of dialogues or meetings. In that case, abstractive summarization tools are needed. Research in that field is very old but has been particularly stimulated since the mid-2010s by the recent successes of deep learning. This paper presents a short survey of deep learning approaches to abstractive text summarization and then highlights the various challenges that will have to be solved in the coming years to deal with meeting summaries in order to be able to provide a text summarization tool that generates good quality summaries.