Location Normalization for Information Extraction

Location Normalization for Information Extraction* Huifeng Li, Rohini K. Srihari, Cheng Niu, and Wei Li Cymfony Inc. 600 Essjay Road, Williamsville, NY 14221, USA (hli, rohini, cniu, wei)@cymfony.com Abstract Ambiguity is very high for location names. For example, there are 23 cities named ‘Buffalo’ in the U.S. Country names such as ‘Canada’, ‘Brazil’ and ‘China’ are also city names in the USA. Almost every city has a Main Street or Broadway. Such ambiguity needs to be handled before we can refer to location names for visualization of related extracted events. This paper presents a hybrid approach for location normalization which combines (i) lexical grammar driven by local context constraints, (ii) graph search for maximum spanning tree and (iii) integration of semi-automatically derived default senses. The focus is on resolving ambiguities for the following types of location names: island, town, city, province, and country. The results are promising with 93.8% accuracy 1 on our test collections. Introduction The task of location normalization is to identify the correct sense of a possibly ambiguous location Named Entity (NE). Ambiguity is very serious for location NEs. For example, there are 23 cities named ‘Buffalo’, including the city in New York State and in Alabama State. Even country names such as ‘Canada’, ‘Brazil’, and ‘China’ are also city names in the USA. Almost every city has a Main Street or Broadway. Such ambiguity needs to be properly handled before converting location names into some normal form to support entity profile construction, event merging and visualization of extracted events on *This work was partly supported by a grant from the Air Force Research Laboratory’s Information Directorate (AFRL/IF), Rome, NY, under contract F30602-00-C-0090. a map for an Information Extraction (IE) System. Location normalization is a special application of word sense disambiguation (WSD). There is considerable research on WSD. Knowledge-based work, such as (Hirst, 1987; McRoy, 1992; Ng and Lee, 1996) used hand-coded rules or supervised machine learning based on annotated corpus to perform WSD. Recent work emphasizes corpus-based unsupervised approach (Dagon and Itai, 1994; Yarowsky, 1992; Yarowsky, 1995) that avoids the need for costly truthed training data. Location normalization is different from general WSD in that the selection restriction often used for WSD in many cases is not sufficient to distinguish the correct sense from the other candidates. For example, in the sentence “The White House is located in Washington”, the selection restriction from the collocation ‘located in’ can only determine that “Washington” should be a location name, but is not sufficient to decide the actual sense of this location. Location normalization depends heavily on co-occurrence constraints of geographically related location entities mentioned in the same discourse. For example, if ‘Buffalo’, ‘Albany’ and ‘Rochester’ are mentioned in the same document, the most probable senses of ‘Buffalo’, ‘Albany’ and ‘Rochester’ should refer to the cities in New York State. There are certain fixed keyword-driven patterns from the local context, which decide the sense of location NEs. These patterns use keywords such as ‘city’, ‘town’, ‘province’, ‘on’, ‘in’ or …

Content truncated to 500 words