Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
コンテクストベクトルの発生と検索のシステム及び方法
Document Type and Number:
Japanese Patent JP4192213
Kind Code:
B2
Abstract:
PROBLEM TO BE SOLVED: To shorten a retrieval time and to improve the availability of the system by applying a learning rule to each word of a record, assigning context vectors according to word closeness, and combining them into a total vector. SOLUTION: The system processes documents of a training set preliminarily (301, 302) and allocates an initial state (303). Then all the documents in a training corpus are processed from the starting document (304), the documents are scanned, one by one, from the starting word trunk as a target (305), and the learning rule is applied to the target (306). After it is confirmed whether or not many trunks are regarded as targets (308), a next trunk is regarded as a target for repetition (308) and when the context vectors of all the trunks are regarded as a target, the total vector of the whole document is determined (309). Further, similar processes are repeated for more documents (310), corrected values are totalized at the time of the repetition to update the vectors in the corpus (312), and the mean is also update (313). Further, it is confirmed whether or not the repetition is required (314) and a total vector is determined and stored (315).

Inventors:
William Robert Cade
Poo Oin
Application Number:
JP12277097A
Publication Date:
December 10, 2008
Filing Date:
April 07, 1997
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
Fair Isaac Corporation
International Classes:
G06F15/18; G06F17/30; G06F17/28; G06N3/00
Domestic Patent References:
JP6195388A
JP7114572A
JP7121549A
Attorney, Agent or Firm:
Minoru Nakamura
Fumiaki Otsuka
Shishido Kaichi
Hideto Takeuchi
Toshio Imajo
Nobuo Ogawa
Village shrine Atsuo