Graph-based Method for Summarized Storyline Generation in Twitter

TL;DR

This paper presents a graph-based framework for generating summarized, coherent storylines from Twitter data by identifying and organizing sub-events, helping users access key information amidst information overload.

Contribution

The paper introduces a novel graph-theoretic approach to extract and organize social salient sub-events into a storyline from Twitter data.

Findings

Effective summarization of Twitter events demonstrated on Iran Election data.

Generated storylines are coherent and highlight key sub-events.

Framework reduces redundancy and improves information accessibility.

Abstract

Twitter has become a leading source of real-time world-wide information and a great medium for exploring emerging events, breaking news and general topics which most matter to a broad audience. On the other hand, the explosive rate of incoming information in Twitter leads users to experience information overload. Whereas, a significant fraction of tweets are about news events, summarizing the storyline of events can be helpful for users to easily access to the relevant and key information hidden among tweets and thereby draw high level conclusions. Storytelling is the task of providing chronological summaries of significant sub-events development and sketching the relationship between sub-events. In this paper, we propose a novel framework to generate a summarized storyline of news events from social point of view. Utilizing the concepts in graph-theory, we identify sub-events, summarize the evolution of sub-events and generate a coherent storyline of them. Our approach models a storyline as a directed tree of social salient sub-events evolving over time. To overcome the enormous number of redundant tweets, we keep distilled information in super-tweets. Experiments performed on a large scale data set from tweets sent during the Iranian Presidential Election (#IranElection) and the results demonstrate the efficiency and effectiveness of our framework.