Harvesting Related Entities with a Search Engine

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1019–1027, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Harvesting Related Entities with a Search Engine∗ Shuqi Sun1, Shiqi Zhao2,1, Muyun Yang1, Haifeng Wang2, and Sheng Li1 1Harbin Institute of Technology, Harbin, China {sqsun,ymy}@mtlab.hit.edu.cn, lisheng@hit.edu.cn 2Baidu, Beijing, China {zhaoshiqi,wanghaifeng}@baidu.com Abstract This paper addresses the problem of re- lated entity extraction and focuses on ex- tracting related persons as a case study. The proposed method builds on a search engine. Specifically, we mine candidate related persons for a query person q us- ing q’s search results and the query logs containing q. The acquired candidates are then automatically rated and ranked using a SVM regression model that investigates multiple features. Experimental results on a set of 200 randomly sampled query per- sons show that the precision of the ex- tracted top-1, 5, and 10 related persons ex- ceeds 91%, 90%, and 84%, respectively, which significantly outperforms a state-of- the-art baseline. 1 Introduction Facilitating efficient navigation in the knowl- edge space is essential to satisfying the current Web search demands. Named entities are vital building blocks of such a space, and retrieving related entities provides an efficient way of nav- igation. Related entity extraction refers to min- ing from text resources named entities with cer- tain relationships between them, e.g. person- affiliation and organization-location. To this end, great efforts have been made recently in both aca- demic (Banko et al., 2007; Wu and Weld, 2010) and industry (Zhu et al., 2009; Shi et al., 2010) circles. A wide range of NLP applications could bene- fit from the high-quality repository of related en- tities. For query suggestion in Web search and e- business, given a query concerning some entity e, one can suggest entities related to e, in which users ∗This work was done when the first author was visiting Baidu. may also have interest. This could be regarded as a remarkable complement to the current techniques suggesting queries that merely contain the entity e or are similar in wording with e (Boldi et al., 2009). In online encyclopedia (e.g. Wikipedia) construc- tion, linking together related entities can facilitate the users for effective navigation. Additionally, re- lated entity extraction also allows us to automati- cally construct social networks. In this paper, we focus on related person extrac- tion, though our proposed techniques can be ex- tended to other types of entities. Here, we pro- vide a comprehensive definition of related persons, which fall into the following four categories. • Persons with definite relationships. The re- lationships in this category can be explicitly represented with definite concepts, e.g. par- ent, friend, colleague, etc. Most previous literature focuses on such definite relation- ships between persons (Brin,1998; Etzioni et al.,2005; Banko et al.,2007; Zhu et al.,2009). • Persons related in certain events. In the second category, the related persons inter- act with the query person in certain events, such as co-starring in the same movie or co- authoring in the same scientific …

Content truncated to 500 words