Tweet Ranking Based on Heterogeneous Networks

Proceedings of COLING 2012: Technical Papers, pages 1239–1256, COLING 2012, Mumbai, December 2012. Tweet Ranking Based on Heterogeneous Networks Hongzhao HUANG1 Arkaitz ZUBIAGA1 Heng JI1 Hong bo DENG2 Dong WANG2 Hieu LE2 Tarek ABDELZAHER2 Jiawei HAN 2 Alice LEUNG3 John HANCOCK4 Clare VOSS5 (1) COMPUTER SCIENCE DEPARTMENT AND LINGUISTICS DEPARTMENT, QUEENS COLLEGE AND GRADUATE CENTER, CITY UNIVERSITY OF NEW YORK, New York, NY, USA 11367 (2) COMPUTER SCIENCE DEPARTMENT, UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN, Urbana-Champaign, IL, USA 61820 (3) RAYTHEON BBN TECHNOLOGIES, Cambridge, MA, USA 02138 (4) ARTISTECH INC., Fairfax, VA, USA 22030 (5) MULTILINGUAL COMPUTING BRANCH, ARMY RESEARCH LAB, Adelphi, MD, USA 20783 {hongzhaohuang, arkaitz.zubiaga, hengjicuny}@gmail.com Abstract Ranking tweets is a fundamental task to make it easier to distill the vast amounts of information shared by users. In this paper, we explore the novel idea of ranking tweets on a topic using heterogeneous networks. We construct heterogeneous networks by harnessing cross-genre linkages between tweets and semantically-related web documents from formal genres, and inferring implicit links between tweets and users. To rank tweets effectively by capturing the semantics and importance of different linkages, we introduce Tri-HITS, a model to iteratively propagate ranking scores across heterogeneous networks. We show that integrating both formal genre and inferred social networks with tweet networks produces a higher-quality ranking than the tweet networks alone. 1 Title and Abstract in Chinese 基 基 基于 于 于异 异 异构 构 构网 网 网络 络 络的 的 的微 微 微信 信 信息 息 息排 排 排序 序 序 微信息排序是一个可以过滤由用户分享的大量信息的根本任务。 在这篇文章中， 我们探索利用异构网络来排序微信息。我们利用来源于正规类型并且与微信息语义相关的 网络文本的跨类型联接, 和推理微信息和用户之间的潜在联系来构造异构网络。为了有效 地捕获不同联接的语义和重要性，我们提出了Tri-HITS, 一个能够跨网络循环传播排序分数 的模型。我们证明了结合来自正规类型的信息，和推理隐含的社交网络可以取得比仅靠微 信息网络本身更高的排序质量。2 Keywords: Tweet Ranking, Heterogeneous Networks, Iterative Propagation Model. Keywords in Chinese: 微信息排序，异构网络，循环传播模型. 1Related resources and software are freely available for research purposes at http://nlp.cs.qc.cuny.edu/tweetranking.zip; the system demo is at http://nlp.cs.qc.cuny.edu/tweet_summary/ground-truth-demo.xhtml. 2有关的资源和程序公布在如下地址和研究相关的应用分享: http://nlp.cs.qc.cuny.edu/tweetranking.zip; 系统演示 在如下地址： http://nlp.cs.qc.cuny.edu/tweet_summary/ground-truth-demo.xhtml. 1239 1 Introduction Twitter has become a popular service for online communication through short messages of up to 140 characters, known as tweets. Its users produce millions of tweets a day, enabling both individuals and organizations to disseminate information about current affairs and breaking news in a timely fashion. This information is sometimes posted by users on-site or in the vicinity of events, providing first-hand accounts from a wide variety of sources. However, the sheer volume of tweets sent during events of general interest is overwhelming and hence difficult to distill for the most relevant information, while also filtering out non-informative tweets. To facilitate finding informative and trustworthy content in tweets, it is crucial to develop an effective ranking method. This is particularly useful in emerging situations. Eyewitnesses might be live-tweeting about anything happening at ongoing events (Diakopoulos et al., 2012) such as natural disasters. To assist in these situations, we aim to develop a ranking system that organizes tweets by informativeness, so that informative tweets are readily identified, while pointless and speculative observations are filtered out. However, the definition of informativeness might vary for different points of view. Twitter users can produce diverse content ranging from news and events, to conversations and personal status updates. …

Content truncated to 500 words