Using Relevant Public Posts to Enhance News Article Summarization

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 557–566, Osaka, Japan, December 11-17 2016. Using Relevant Public Posts to Enhance News Article Summarization Chen Li1, Zhongyu Wei2∗, Yang Liu3, Yang Jin3, Fei Huang4 1 Microsoft, Bellevue, WA, USA 2 School of Data Science, Fudan University, Shanghai, P.R.China 3 Computer Science Department, The University of Texas at Dallas 4 Facebook, 770 Broadway, New York, NY, USA chei@microsoft.com zywei@fudan.edu.cn {yangl,yangjin@hlt.utdallas.edu} feihuang@fb.com Abstract A news article summary usually consists of 2-3 key sentences that reflect the gist of that news article. In this paper we explore using public posts following a new article to improve automatic summary generation for the news article. We propose different approaches to incorporate infor- mation from public posts, including using frequency information from the posts to re-estimate bigram weights in the ILP-based summarization model and to re-weight a dependency tree edge’s importance for sentence compression, directly selecting sentences from posts as the final sum- mary, and finally a strategy to combine the summarization results generated from news articles and posts. Our experiments on data collected from Facebook show that relevant public posts pro- vide useful information and can be effectively leveraged to improve news article summarization results. 1 Introduction Nowadays people are often overwhelmed by their daily exposure to large amount of online information. To make information easier to digest, news press like CNN, USAToday or news disseminator like Yahoo often provide ‘summaries’ for their news articles, so that readers can get the gist of a story quickly. Typically this kind of short summaries is manually generated. Obviously, it is very time consuming to manually produce high quality summaries for many popular topics. Therefore, automatic summarization for related news articles is essential to alleviate the manual work. With the popularity of social media, online news providers or disseminators are moving towards offering more interactions with news readers, for example, via comments on the news provide sites or post service like Twitter or Facebook public posts. When a news is published, we have access to not only the related news articles, but also the related public comments and posts. Our task in this paper is thus to explore how to use relevant public posts to improve summarization of a single news article. In particular, we use Facebook public posts related to a news article to help summarize a popular topic. This work is also motivated by the following observations of the data (see Sec 3 for the data we use). First, the posts under a news article are closely related to and very indicative for the topic of that news story. Second, the sentences from some posts whose accounts are maintained by news agencies are well written, so they may be directly used as the units of extractive summarization. In addition, the sentences in posts are often shorter than those from the news, thus again they may be more suitable to be used as summary sentences in sentence-based extractive summarization. Our contributions in …

Content truncated to 500 words