Creator |
|
|
Language |
|
Publisher |
|
|
Date |
|
Source Title |
|
|
Vol |
|
Issue |
|
First Page |
|
Last Page |
|
Conference |
|
Publication Type |
|
Access Rights |
|
Related DOI |
|
Related DOI |
|
Related URI |
|
Related URI |
|
Related HDL |
|
Relation |
|
Abstract |
This paper proposes a method which constructs a hierarchy of words from given a set of documents automatically and dynamically. The hierarchy of the words is constructed as a hypernym relation that is... defined by the document frequencies and co-occurrence probability of words. The hierarchy is obtained not only to the whole set of documents, but also to any subset of documents. A typical example of such documents is the search results of a keyword. The hierarchy obtained to this set is the hierarchy of related words of the keywords. Empirical evaluations are conducted for the word hierarchy derived from "Eijiro", an English-Japanese dictionary which contains 1,648,628 descriptions of words. The hierarchies are compared with the proposed method, Niwa's method and Shrinivasan's method with respect to coverage and granularity. This paper proposes a method which constructs a hierarchy of words from given a set of documents automatically and dynamically. The hierarchy of the words is constructed as a hypernym relation that is defined by the document frequencies and co-occurrence probability of words. The hierarchy is obtained not only to the whole set of documents, but also to any subset of documents. A typical example of such documents is the search results of a keyword. The hierarchy obtained to this set is the hierarchy of related words of the keywords. Empirical evaluations are conducted for the word hierarchy derived from "Eijiro", an English-Japanese dictionary which contains 1,648,628 descriptions of words. The hierarchies are compared with the proposed method, Niwa's method and Shrinivasan's method with respect to coverage and granularity.show more
|