A Data Driven Approach for Person Name Disambiguation in Web Search Results

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 301–310, Dublin, Ireland, August 23-29 2014. A Data Driven Approach for Person Name Disambiguation in Web Search Results Agust´ın D. Delgado1, Raquel Mart´ınez1, V´ıctor Fresno1, Soto Montalvo2 1Universidad Nacional de Educaci´on a Distancia (UNED), Madrid, Spain 2Universidad Rey Juan Carlos (URJC), M´ostoles, Spain 1{agustin.delgado,raquel,vfresno}@lsi.uned.es, 2soto.montalvo@urjc.es Abstract This paper presents an unsupervised approach for the task of clustering the results of a search engine when the query is a person name shared by different individuals. We propose an algo- rithm that calculates the number of clusters and establishes the groups of web pages according to the different individuals without the need to any training data or predefined thresholds, as the successful state of the art systems do. In addition, most of those systems do not deal with social media web pages and their performance could fail in a real scenario. In this paper we also propose a heuristic method for the treatment of social networking profiles. Our approach is compared with four gold standard collections for this task obtaining really competitive results, comparable to those obtained by some approaches with supervision. 1 Introduction Resolving the ambiguity of person names in web search results is a challenging problem becoming an area of interest for Natural Language Processing (NLP) and Information Retrieval (IR) communities. This task can be defined informally as follows: given a query of a person name in addition to the results of a search engine for that query, the goal is to cluster the resultant web pages according to the different individuals they refer to. Thus, the challenge of this task is estimating the number of different individuals and grouping the pages of the same individual in the same cluster. The difficulty of this task resides in the fact that a single person name can be shared by many people: according to the U.S. Census Bureau, 90000 different names are shared by 100 million people (Artiles et al., 2007). This problem has had an impact in the Internet and that is why several vertical search engines specialized in web people search have appeared in the last years, e.g. spokeo.com or 123people.com. This task should not be mixed up with entity linking (EL), which goal is to link name mentions of entities in a document collection to entities in a reference knowledge base (typically Wikipedia), or to detect new entities. The main difficulties of clustering web pages referring to the same individual come from their possible heterogeneous nature. For example, some pages may be professional sites, while others may be blogs containing personal information. In addition, the popularity of social networking services makes the search engine usually returns several social profiles belonging to different individuals sharing the same name, as much from the same social networking service as from different services. These social pages often introduce noisy information and make the state of the art algorithms break down (Berendsen et al., 2012). Due to these problems, the …

Content truncated to 500 words