Method for predicting negative example, system for detecting incorrect wording using negative example prediction
First Claim
1. A method for predicting whether data that is not yet known to be positive or negative with respect to a certain problem is positive or negative, comprising:
- a positive example accessing step of accessing a positive example data storage unit pre-storing groups of positive example data constituting correction examples for the problem;
an existence determination step for determining whether or not the data exists in the positive example data groups;
an appearance probability calculating step for calculating typical probability of appearance of the data when the data does not exist in the positive example data groups; and
an negative example likelihood calculation step for calculating a probability of appearance of the data in the positive example data group based on the typical probability of appearance and taking the probability as a likelihood of an negative example.
2 Assignments
0 Petitions
Accused Products
Abstract
An negative example prediction processing method for predicting a likelihood of examples being negative for data where, with respect to a certain problem, it is not known whether the data is for a correctly worded positive example or for an incorrectly worded negative example. In this negative example prediction processing method, an unknown example x is inputted and a determination is made as to whether or not the example x exists in a positive example database provided in advance. If the example x does not exist, a typical probability of appearance p(x) for the example x is calculated, and a likelihood Q(x) of the example x being an negative example is calculated from the probability of appearance p(x).
28 Citations
13 Claims
-
1. A method for predicting whether data that is not yet known to be positive or negative with respect to a certain problem is positive or negative, comprising:
-
a positive example accessing step of accessing a positive example data storage unit pre-storing groups of positive example data constituting correction examples for the problem;
an existence determination step for determining whether or not the data exists in the positive example data groups;
an appearance probability calculating step for calculating typical probability of appearance of the data when the data does not exist in the positive example data groups; and
an negative example likelihood calculation step for calculating a probability of appearance of the data in the positive example data group based on the typical probability of appearance and taking the probability as a likelihood of an negative example. - View Dependent Claims (2, 3)
-
-
4. A recording medium recorded with a program for implementing a processing method for predicting whether data that is not yet known to be positive or negative with respect to a certain problem is positive or negative on a computer, comprising:
-
a positive example accessing step of accessing a positive example data storage unit pre-storing groups of positive example data constituting correction examples for the problem;
an existence determination step for determining whether or not the data exists in the positive example data groups;
an appearance probability calculating step for calculating typical probability of appearance of the data when the data does not exist in the positive example data groups; and
an negative example likelihood calculation step for calculating a probability of appearance of the data in the positive example data group based on the typical probability of appearance and taking the probability as a likelihood of an negative example. - View Dependent Claims (5, 6)
-
-
7. A recording medium recorded with a program for implementing processing to detect incorrect wording using a method for predicting negative examples on a computer, comprising:
-
a positive example data accessing step of accessing a positive example data storage unit pre-storing groups of positive example data constituting correctly worded data;
an existence determination step for determining whether or not the inputted wording exists in the positive example data groups;
an appearance probability calculating step for calculating typical probability of appearance of the inputted wording when inputted wording data does not exist in the positive example data groups; and
an negative example likelihood calculation step for calculating a probability of appearance of the inputted wording in the positive example data group based on the typical probability of appearance and taking the probability as a likelihood of an negative example. - View Dependent Claims (8)
-
-
9. A recording medium recorded with a program for implementing processing to extract a embedded clause constituting a non-case relational relative clause using a method for predicting negative examples on a computer, comprising:
-
a positive example data accessing step of accessing a positive example data storage unit pre-storing groups of positive example data constituting case relational relative clauses;
an existence determination step for determining whether or not the inputted embedded example exists in the positive example data group;
an appearance probability calculating step for calculating typical probability of appearance of the embedded clause when the embedded clause data does not exist in the positive example data groups; and
an negative example likelihood calculation step for calculating a probability of appearance of the embedded clause in the positive example data group based on the typical probability of appearance and taking the probability as a likelihood of an negative example.
-
-
10. A method for detecting incorrect wording using a supervised machine learning method, comprising the steps of:
-
extracting pairs of feature and solutions from supervised data including correctly worded positive example data and incorrectly worded negative example data;
performing machine learning taking pairs of extracted features and solutions as borrowing supervised data and storing learning results in a learning results data storage unit; and
extracting features from inputted data and detecting incorrect wording based on the learning results saved in the learning results data storage unit.
-
-
11. A recording medium recorded with a program for implementing processing to detect incorrect wording using a supervised machine learning method, comprising the steps of:
-
processing for extracting pairs of features and solutions from supervised data including correctly worded positive example data and incorrectly worded negative example data;
performing machine learning processing taking pairs of features and solutions as borrowing supervised data and storing learning results in a learning results data storage unit; and
processing for extracting features from inputted data and detecting incorrect wording in the data based on the learning results.
-
-
12. A processing method for detecting incorrect wording comprising the steps of:
-
calculating typical probability of appearance for examples when inputted examples to not exist in prepared correctly worded positive example data;
calculating probability of the example appearing in the positive example data based on the typical probability of appearance, and taking the example as negative example data when the probability exceeds a prescribed threshold value;
extracting pairs of features and solutions from supervised data including positive example data and negative example data;
performing machine learning processing taking pairs of features and solutions as borrowing supervised data, and storing learning results in a learning results data storage unit; and
extracting features from inputted data and detecting incorrect wording in the data based on the learning results.
-
-
13. A recording medium recorded with a program for implementing processing to detect incorrect wording using supervised machine learning methods, comprising the steps of:
-
calculating typical probability of appearance for examples when inputted examples to not exist in prepared correctly worded positive example data;
calculating probability of the example appearing in the positive example data based on the typical probability of appearance, and taking the example as negative example data when the probability exceeds a prescribed threshold value;
extracting pairs of features and solutions from supervised data including positive example data and negative example data;
performing machine learning processing taking pairs of features and solutions as borrowing supervised data, and storing learning results in a learning results data storage unit; and
extracting features from inputted data and detecting incorrect wording in the data based on the learning results.
-
Specification