ES-LDA: Entity Summarization using Knowledge-based Topic Modeling

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 316–325, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP ES-LDA: Entity Summarization using Knowledge-based Topic Modeling Seyedamin Pouriyeh1, Mehdi Allahyari∗2, Krys Kochut1, Gong Cheng3, and Hamid Reza Arabnia1 1Computer Science Department, University of Georgia, Athens, GA, USA {pouriyeh,kkochut,hra@uga.edu} 2Department of Computer Science, Georgia Sothern University, Statesboro, USA {mallahyari@georgiasouthern.edu} 3National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China {gcheng@nju.edu.cn} Abstract With the advent of the Internet, the amount of Semantic Web documents that describe real-world entities and their inter-links as a set of statements have grown consid- erably. These descriptions are usually lengthy, which makes the utilization of the underlying entities a difficult task. En- tity summarization, which aims to cre- ate summaries for real world entities, has gained increasing attention in recent years. In this paper, we propose a probabilistic topic model, ES-LDA, that combines prior knowledge with statistical learning tech- niques within a single framework to cre- ate more reliable and representative sum- maries for entities. We demonstrate the effectiveness of our approach by conduct- ing extensive experiments and show that our model outperforms the state-of-the-art techniques and enhances the quality of the entity summaries. 1 Introduction With the emergence of Linked Open Data (LOD)1 as a way of publishing and interacting with the in- formation, many datasets such as DBpedia (Bizer et al., 2009) and YAGO (Hoffart et al., 2013) have been created and are publicly available on the Web. For example, DBpedia as part of LOD is a knowledge base extracted from Wikipedia that consists of Wikipedia resources (entities) described as RDF statements (i.e., RDF triples). The Resource Description Framework (RDF) is the Semantic Web standard data model used for representing information on the Web. An RDF triple is represented in the form of ∗Equal contribution 1http://linkeddata.org < subject, predicate, object >. The latest En- glish version of DBpedia contains over 4.5 million entities collectively described by over 1.6 billion triples. This means that each entity description has an average of 355 RDF triples. Human users and computer applications need to consider these lengthy descriptions while performing various se- mantic tasks. Thus, entity summarization, a task of producing more concise, but still sufficient en- tity description, has garnered a significant amount of attention. Recently, with the huge growth of information, summarization techniques are becoming some of the main approaches to making the information more readily available. In fact, summarization techniques aim to facilitate the identification of structure and meaning in data. Researchers in different communities have taken a strong inter- est in this task and, accordingly, have proposed various methods for a wide variety of summa- rization techniques in multiple areas. Document summarization (Nenkova and McKeown, 2012), database summarization (Bu et al., 2005), and graph summarization (Navlakha et al., 2008) are just a few examples that have been studied by different communities. RDF data summarization and in particular entity summarization, has at- tracted considerable attentions in recent years as it …

Content truncated to 500 words