Reputation information processing program, method, and apparatus

A reputation information processing method, comprising:

• an evaluation pair extraction step of analyzing a document which is present on a network, extracting an evaluation pair in which an object and an evaluation expression are combined, and registering the evaluation pair in a database;
• and
• an evaluation pair analysis step of executing desired analysis by utilizing the evaluation pair of the database, wherein
• the evaluation pair extraction step further comprises:
• an evaluation pair learning step of giving an evaluation pair extracted from a model document for training to a machine learner so as to cause the machine learner to learn a rule which can serve as the evaluation pair;
• and
• an evaluation pair extraction execution step of inputting an unknown document to the machine learner and extracting an evaluation pair based on the learnt rules,
• the evaluation pair learning step includes:
• a morpheme analysis step of subjecting the model document to morpheme analysis when a word is in a language in which words are continuous without being punctuated, and skipping the morpheme analysis of said model document when a word is in a language in which words are punctuated by spaces;
• an object extraction step of extracting a unique expression and a noun phrase from the model document which has undergone morpheme analysis and causing the noun expression and the noun phrase to serve as objects;
• an evaluation expression extraction step of extracting an evaluation expression from the model document which has undergone morpheme analysis by use of an evaluation expression dictionary;
• a learning evaluation pair extraction step of extracting an evaluation pair candidate in which the object and the evaluation expression are combined, extracting a feature amount of the evaluation pair candidate, and setting whether the combination of the evaluation pair candidate is correct or not;
• and
• a model generating step of inputting a plurality of the feature amounts and the correctness of the learning evaluation pair to the machine learner, and generating weights of the individual feature amounts as an evaluation pair sorting model in which the rule which can serve as the learning evaluation pair is learnt and acquired, and
• the evaluation pair extraction execution step includes:
• a morpheme analysis step of subjecting an unknown document to morpheme analysis in a case of a language in which words are continuous without being punctuated by spaces, and skipping a morpheme analysis of said model document in a case of a language in which words are punctuated by spaces;
• an object extraction step of extracting a unique expression and a noun phrase from the unknown document which has undergone morpheme analysis and causing the noun expression and the noun phrase to serve as objects;
• an evaluation expression extraction step of extracting an evaluation expression from the unknown document which has undergone morpheme analysis by use of an evaluation expression dictionary;
• an evaluation pair candidate extraction step of extracting an evaluation pair candidate in which the object and the evaluation expression are combined;
• and
• an evaluation pair determination step of inputting the evaluation pair candidate to the machine learner, giving a plurality of feature amounts to the evaluation pair candidate by use of the evaluation pair sorting model, determining the evaluation pair candidate having a total sum of the individual feature amounts multiplied by the weights acquired in the evaluation pair learning step, equal to or more than a predetermined value as an evaluation pair of the unknown document, and registering the evaluation pair in the database.

A reputation information processing apparatus embodied as a processor, the apparatus, comprising:

• an evaluation pair extraction unit analyzing a document which is present on a network, extracting an evaluation pair composed of an object and an evaluation expression, and registering the evaluation pair in a database;
• and
• an evaluation pair analysis unit executing desired analysis by utilizing the evaluation pair of the database, wherein
• the evaluation pair extraction unit performs an evaluation pair learning step of giving an evaluation pair extracted from a model document for training to a machine learner so as to cause the machine learner to learn a rule which can serve as the evaluation pair and an evaluation pair extraction execution step of inputting an unknown document to the machine learner and extracts an evaluation pair based on the learnt rules,
• the evaluation pair learning step includes:
• a morpheme analysis step of subjecting the model document to morpheme analysis when a word is in a language in which words are continuous without being punctuated, and skipping the morpheme analysis of said model document when a word is in a language in which words are punctuated by spaces;
• an object extraction step of extracting a unique expression and a noun phrase from the model document which has undergone morpheme analysis and causing the noun expression and the noun phrase to serve as objects;
• an evaluation expression extraction step of extracting an evaluation expression from the model document which has undergone morpheme analysis by use of an evaluation expression dictionary;
• a learning evaluation pair extraction step of extracting an evaluation pair candidate in which the object and the evaluation expression are combined, extracting a feature amount of the evaluation pair candidate, and setting whether the combination of the evaluation pair candidate is correct or not;
• and
• a model generating step of inputting a plurality of the feature amounts and the correctness of the learning evaluation pair to the machine learner, and generating weights of the individual feature amounts as an evaluation pair sorting model in which the rule which can serve as the learning evaluation pair is learnt and acquired, and
• the evaluation pair extraction execution step includes:
• a morpheme analysis step of subjecting an unknown document to morpheme analysis in a case of a language in which words are continuous without being punctuated by spaces, and skipping a morpheme analysis of said model document in a case of a language in which words are punctuated by spaces;
• an object extraction step of extracting a unique expression and a noun phrase from the unknown document which has undergone morpheme analysis and causing the noun expression and the noun phrase to serve as objects;
• an evaluation expression extraction step of extracting an evaluation expression from the unknown document which has undergone morpheme analysis by use of an evaluation expression dictionary;
• an evaluation pair candidate extraction step of extracting an evaluation pair candidate in which the object and the evaluation expression are combined;
• and
• an evaluation pair determination step of inputting the evaluation pair candidate to the machine learner, giving a plurality of feature amounts to the evaluation pair candidate by use of the evaluation pair sorting model, determining the evaluation pair candidate having a total sum of the individual feature amounts multiplied by the weights acquired in the evaluation pair learning step, equal to or more than a predetermined value as an evaluation pair of the unknown document, and registering the evaluation pair in the database.