Summarization Systems: From Text to Multimodal

Abstract: Text summarization refers to generating a crisp and concise description of a text by ignoring irrelevant and noisy content and focusing on important and relevant content and it is one of the key research areas of natural language processing. It has several real-life applications starting from news summarization, microblog summarization, scientific document summarization, writing literature reviews, etc. If the generated summary is a subset of the original text document, then this is called extractive summarization whereas if new sentences are generated as a part of summary, then this is termed as abstractive summarization. In today's scenario, due to the increase in the use of social media, a large number of data are being generated continuously. To deal with such data, the summary generation technique can be adapted. Sharing information via the Internet has become the most popular means of transferring information. Thanks to technological advancements, people have effective means to share content in multi-media formats. However, the vast plethora of public opinion and facts on a topic makes it difficult to access the crux of a topic. Prior studies have shown that multi-modal output containing images and text increases user satisfaction by 12.4% as compared to single modal output such as text only summary. Thus recent studies have focused on developing some multimodal summarization techniques where output summary may contain some text summary, image summary and video summary.

In this tutorial, different types of summarization in detail, extractive, and abstractive will be discussed. Different classical optimization-based approaches for extractive summarization will be presented followed by recently introduced evolutionary based approaches. The tutorial will also cover recently introduced multimodal summarization techniques and will showcase with examples how multimodality helps in generating better quality summary. This tutorial will discuss in detail various application domains where summarization techniques can be applied including news summarization, micro-blog summarization, multi-modal summarization, scientific document summarization etc. Throughout the tutorial we will consider various text and multi-modal data sets and will try to compare different text summary generation approaches.

Outline of the Tutorial:

  • What is summarization and different types of summarization with examples (5 Min)
  • Why Summarization (5 min)
  • Application areas of summarization (5 min)
  • Data sets and evaluation metrics for text summarization (10 min)
  • Approaches to text summarization (20 min)
  • Different approaches for multimodal summarization (10 min)
  • Microblog summarization (5 min)
  • Scientific document summarization (5 min)
  • Deep-based abstractive text summarization methods (static data) (25 min)
  • Streaming data text summarization methods (15 min)
  • Multimodal Microblog summarization: concept and utility (10 min)
  • Figure-associated Text Summarization (10 min)
  • Comment-oriented Document Summarization (15 Min)
  • Hands-on Session (1:30 Hour)
  • Future scope of summarization (10 min)
  • Conclusion (5 min)

Level of the Tutorial: Introduction to advanced level.

Expected Length of the Tutorial: Half day (Approx. 4 hours)

Any Hands-on Session: Yes

Target Audience: We are targeting the audience of bachelor students, Masters students and researchers. The tutorial will also be beneficial to those who want to do some research in the summarization domain. The audience with basic machine learning and deep learning knowledge will be able to understand.


Dr. Naveen Saini, Assistant Professor, Department of Computer Science, Indian Institute of Information Technology Lucknow
Email ID:;

Expertise: Automatic Summarization, Data Mining, Machine Learning, Multi-objective Optimization, and Evolutionary Algorithms

Biography: Dr. Naveen Saini is an Assistant Professor in the Department of Computer Science at Indian Institute of Information Technology Lucknow. Before this, he was working as a researcher at 4IR Applied Research Center and Assistant Professor at Endicott College of International Studies, Woosong University, South Korea. He did his post doctorate from IRIT (Institut De Recherche En Informatique De Toulouse) which is a joint research unit of Université Toulouse III - Paul Sabatier, Toulouse, France. During postdoctoral, he was blessed to work with Dr. Jose Moreno and Prof. Antoine Doucet. He earned my PhD in the Department of Computer Science and Engineering at Indian Institute of Technology Patna (IITP), India under the guidance of Dr. Sriparna Saha and Prof. Pushpak Bhattacharya. His current research interests include developing unsupervised algorithms for Text Clustering, Automatic Summarization Systems, Scholarly Data Mining using Machine Learning, Multi-objective Optimization, and Evolutionary Algorithms. His more information can be found at