Generating Templates of Entity Summaries with an Entity-Aspect Model and Pattern Mining

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 640–649, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Generating Templates of Entity Summaries with an Entity-Aspect Model and Pattern Mining Peng Li1 and Jing Jiang2 and Yinglin Wang1 1Department of Computer Science and Engineering, Shanghai Jiao Tong University 2School of Information Systems, Singapore Management University {lipeng,ylwang}@sjtu.edu.cn jingjiang@smu.edu.sg Abstract In this paper, we propose a novel approach to automatic generation of summary tem- plates from given collections of summary articles. This kind of summary templates can be useful in various applications. We first develop an entity-aspect LDA model to simultaneously cluster both sentences and words into aspects. We then apply fre- quent subtree pattern mining on the depen- dency parse trees of the clustered and la- beled sentences to discover sentence pat- terns that well represent the aspects. Key features of our method include automatic grouping of semantically related sentence patterns and automatic identification of template slots that need to be filled in. We apply our method on five Wikipedia entity categories and compare our method with two baseline methods. Both quantitative evaluation based on human judgment and qualitative comparison demonstrate the ef- fectiveness and advantages of our method. 1 Introduction In this paper, we study the task of automatically generating templates for entity summaries. An en- tity summary is a short document that gives the most important facts about an entity. In Wikipedia, for instance, most articles have an introduction section that summarizes the subject entity before the table of contents and other elaborate sections. These introduction sections are examples of en- tity summaries we consider. Summaries of enti- ties from the same category usually share some common structure. For example, biographies of physicists usually contain facts about the national- ity, educational background, affiliation and major contributions of the physicist, whereas introduc- tions of companies usually list information such as the industry, founder and headquarter of the company. Our goal is to automatically construct a summary template that outlines the most salient types of facts for an entity category, given a col- lection of entity summaries from this category. Such kind of summary templates can be very useful in many applications. First of all, they can uncover the underlying structures of summary articles and help better organize the information units, much in the same way as infoboxes do in Wikipedia. In fact, automatic template genera- tion provides a solution to induction of infobox structures, which are still highly incomplete in Wikipedia (Wu and Weld, 2007). A template can also serve as a starting point for human edi- tors to create new summary articles. Furthermore, with summary templates, we can potentially ap- ply information retrieval and extraction techniques to construct summaries for new entities automati- cally on the fly, improving the user experience for search engine and question answering systems. Despite its usefulness, the problem has not been well studied. The most relevant work is by Fila- tova et al. (2006) on automatic creation of …

Content truncated to 500 words