Google based name search: Resolving mixed entities on the web

Abstract

When non-unique values are used as the identifier of entities, due to their homonym, confusion can occur. In particular, when part of “names” of entities are used as their identifiers, the problem is often referred to as a mixed entity resolution problem, where goal is to sort out the erroneous entities due to name homonyms (e.g., if only last name is used as an identifier, one cannot distinguish “Vannevar Bush” from “George Bush”). Especially, a mixed entity resolution problem is common on the Web data. For instance, to search for a product name (e.g., Oracle) in Google, there exist a mixture of web pages due to the name homonyms (e.g., Oracle Database, Oracle Audio, Oracle Academy, etc.). In this paper, we present a practical system for resolving such mixed entities on the Web. For development of such a system, we propose a web service based interface, an unsu-pervised clustering scheme, and cluster ranking algorithms. In particular, since the correct number of clusters is often unknown, we study a state-of-the-art unsupervised clustering solution based on propagation of pairwise similarities of entities. Our claim is empirically validated via experimentation, showing that our approach outperforms main competing solution.