| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To allow various analysis of reputation information by extracting reputation information with a high precision in a wide range and to allow attributes of an object to be automatically extracted from evaluation expressions. <P>SOLUTION: An evaluation pair extraction part 18 analyzes a document existing on a network to extract evaluation pairs consisting of objects and evaluation expressions and registers the evaluation pairs in an evaluation pair database 20. An evaluation pair analysis part 24 uses the evaluation pairs in the database to perform category estimation, peculiar evaluation expression extraction, analysis like comparative analysis. The evaluation pairs are data being combinations of objects and evaluation expressions, and the objects are specific objects including trade names, maker names, etc. which are capable of being evaluated, and evaluation expressions of evaluation pairs are words and phrases expressing evaluations including liking, unfriendliness, etc. The evaluation pair extraction part 18 gives evaluation pairs extracted from a model document to a machine learning device to cause the device to learn rules capable of becoming evaluation pairs and then correctly combines objects and evaluation expressions to extract evaluation pairs from an unknown document. <P>COPYRIGHT: (C)2007,JPO&INPIT</p> |
