|
JAIST Repository >
School of Knowledge Science >
Articles >
Journal Articles >
Please use this identifier to cite or link to this item:
http://hdl.handle.net/10119/5018
|
| Title: | An evidential reasoning approach to weighted combination of classifiers for word sense disambiguation |
| Authors: | Le, Cuong Anh
Huynh, Van-Nam
Shimazu, Akira |
| Keywords: | Computational linguistics
Weighted combination of classifiers
Word sense disambiguation
Dempster-Shafer theory of evidence |
| Issue Date: | 2005 |
| Publisher: | Springer |
| Magazine name: | Lecture Notes in Computer Science |
| Volume: | 3587 |
| Start page: | 516 |
| End page: | 525 |
| DOI: | 10.1007/11510888_51 |
| Abstract: | Arguing that various ways of using context in word sense disambiguation (WSD) can be considered as distinct representations of a polysemous word, a theoretical framework for the weighted combination of soft decisions generated by experts employing these distinct representations is proposed in this paper. Essentially, this approach is based on the Dempster-Shafer theory of evidence. By taking the confidence of individual classifiers into account, a general rule of weighted combination for classifiers is formulated, and then two particular combination schemes are derived. These proposed strategies are experimentally tested on the datasets for four polysemous words, namely interest, line, serve, and hard, and obtained the better result in comparison with previous studies for all cases, with an exception of the word line. |
| Rights: | This is the author-created version of Springer, Cuong Anh Le, Van-Nam Huynh and Akira Shimazu, Lecture Notes in Computer Science, 3587, 2005, 516-525. The original publication is available at www.springerlink.com, http://dx.doi.org/10.1007/11510888_51 |
| URI: | http://hdl.handle.net/10119/5018 |
| Material Type: | author |
| Appears in Collections: | a10-1. 雑誌掲載論文 (Journal Articles)
|
Files in This Item:
| File |
Description |
Size | Format |
|---|
| C4760.pdf | | 191Kb | Adobe PDF | View/Open |
|
All items in DSpace are protected by copyright, with all rights reserved.
|
