Research Area:  Machine Learning
Recent advances in microblog content summarization has primarily viewed this task in the context of traditional multi-document summarization techniques where a microblog post or their collection form one document. While these techniques already facilitate information aggregation, categorization and visualization of microblog posts, they fall short in two aspects: i) when summarizing a certain topic from microblog content, not all existing techniques take topic polarity into account. This is an important consideration in that the summarization of a topic should cover all aspects of the topic and hence taking polarity into account (sentiment) can lead to the inclusion of the less popular polarity in the summarization process. ii) Some summarization techniques produce summaries at the topic level. However, it is possible that a given topic can have more than one important aspect that need to have representation in the summarization process. Our work in this paper addresses these two challenges by considering both topic sentiments and topic aspects in tandem. We compare our work with the state of the art Twitter summarization techniques and show that our method is able to outperform existing methods on standard metrics such as ROUGE-1.
Author(s) Name:  Syed Muhammad Ali, Zeinab Noorian, Ebrahim Bagheri, Chen Ding & Feras Al-Obeidat
Journal name:  Journal of Intelligent Information Systems
Publisher name:  Springer
Volume Information:  volume 54, pages 129–156 (2020)
Paper Link:   https://link.springer.com/article/10.1007/s10844-018-0521-8