Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/7432
Title: Named entity disambiguation from web text
Authors: Xu, Jian
Keywords: Text processing (Computer science)
Natural language processing (Computer science)
Names.
Hong Kong Polytechnic University -- Dissertations
Issue Date: 2014
Publisher: The Hong Kong Polytechnic University
Abstract: Named entity disambiguation is the problem of grouping name mentions into clusters, with each cluster referring to the same underlying entity. In this thesis, we focus on named entity disambiguation from web text, because finding information about person on the Internet is one of the most common activities of online users. Person's names, however, are highly ambiguous with a large number of people sharing the same name. Named entity disambiguation therefore becomes increasingly important for many applications such as information retrieval, question answering, cross-document co-reference, relation discovery and so on. This leads to our study of named entity disambiguation over the Internet. In general, named entity disambiguation for web text includes two tasks: (1) Web Person Disambiguation (WPD), which groups search results into different clusters with each cluster referring to the same person; and (2) personal profile extraction (PPE), which can help build each person's relational information in the cluster. The main challenges in named entity disambiguation include (1) how to select meaningful features that are unique to identify named entities; (2) how to guarantee high performance in WPD, even if there is no prior knowledge of the number of persons having the same name; (3) how to obtain and select quality training data from an external knowledge base for personal profile extraction (PPE), since manually annotated data is costly to yield and limited in scale. In this thesis, we explore the use of more semantically relevant information for named entity disambiguation on web text. For WPD, our supervised approach can make good use of naturally annotated resource, Wikipedia in particular to alleviate manual annotation efforts and domain dependence problems. We also investigate the usage of keywords as semantically more meaningful information units for WPD. Based on meaningful keyword features, we investigate a hierarchical co-reference resolution technique to place ambiguous person names into different clusters. Our disambiguation method does not require a predefined number of persons and can produce good quality clusters for each person. For PPE, we build a personalized profile by identifying relational facts. Our approach is to incorporate two semantic constraints, including both trigger word and entity type which can help reduce noisy data in profile extraction. Both WPD and PPE are built within the framework of graphical models, which can provide sequential structure for semantic feature extraction and tree structure for both name disambiguation and profile extraction. The methods in this thesis are evaluated on publicly available datasets so that performance comparisons can be made to state-of-the-art works and our approach is proven to be effective in named entity disambiguation.
Description: xii, 148 pages : color illustrations ; 30 cm
PolyU Library Call No.: [THS] LG51 .H577P COMP 2014 Xu
URI: http://hdl.handle.net/10397/7432
Rights: All rights reserved.
Appears in Collections:Thesis

Files in This Item:
File Description SizeFormat 
b27630018_link.htmFor PolyU Users203 BHTMLView/Open
b27630018_ir.pdfFor All Users (Non-printable)3.5 MBAdobe PDFView/Open
Show full item record

Page view(s)

168
Last Week
4
Last month
Checked on Jan 15, 2017

Download(s)

105
Checked on Jan 15, 2017

Google ScholarTM

Check



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.