Summarizing Complex Events: a Cross-Modal Solution of Storylines Extraction and Reconstruction

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1281–1291, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Summarizing Complex Events: a Cross-modal Solution of Storylines Extraction and Reconstruction Shize Xu xsz@pku.edu.cn Shanshan Wang cheers echo mch@163.com Yan Zhang∗ zhy@cis.pku.edu.cn Department of Machine Intelligence, Peking University, Beijing, China Key Laboratory on Machine Perception, Ministry of Education, Beijing, China Abstract The rapid development of Web2.0 leads to significant information redundancy. Espe- cially for a complex news event, it is diffi- cult to understand its general idea within a single coherent picture. A complex event of- ten contains branches, intertwining narratives and side news which are all called storylines. In this paper, we propose a novel solution to tackle the challenging problem of storylines extraction and reconstruction. Specifically, we first investigate two requisite properties of an ideal storyline. Then a unified algorithm is devised to extract all effective storylines by optimizing these properties at the same time. Finally, we reconstruct all extracted lines and generate the high-quality story map. Exper- iments on real-world datasets show that our method is quite efficient and highly compet- itive, which can bring about quicker, clearer and deeper comprehension to readers. 1 Introduction News reports usually consist of various modalities of tremendous information, especially all kinds of textual information and visual information, which make web users dazzled and lost. The situation gets worse on complex news events. To help readers quickly grasp the general information of the news, a more concise and convenient system over multi- modality information should be provided. For ex- ample, given a large collection of texts and images related to a specified news event (e.g., East Japan ∗Corresponding author Earthquake), such a system should present a terse and brief summarization about the event by showing different clues of its development, and thus helping readers to effectively find out “when, where, what, how and why” at a glance. The researches (Goldstein et al., 2000) on auto- matic multi-document summarization (MDS) have helped a lot when we generate a description for a specific event. However, it traditionally exhibits in a very simple style like a “0-dimensional” point. The appearance of Timeline (Allan et al., 2001) brings about a visual progress for massive documents anal- yses. Readers can not only get the most important ideas, but also browse the story evolution in chrono- logical order. Previous news summarization systems with structured output (Yan et al., 2011) have fo- cused on timeline generation. Timeline becomes a “1-dimensional” line. This style of summarization only works for simple stories, which are linear in na- ture. However, the structure of complex stories usu- ally turns out to be non-linear. These stories branch into storylines, dead ends, intertwining narratives and side news. To explore these lines, we need a map to reorganize all the information. Therefore, a “2-dimensional” story map is in bad need. Fig- ure 1 shows a part of the story map generated by our system for representing East Japan Earthquake. We …

Content truncated to 500 words