摘要 |
PROBLEM TO BE SOLVED: To generate a temporal positive example, and to advance active learning when any positive example does not exist, or rarely exists. SOLUTION: A learning data storage part 131 stores the group of learning data configured of a plurality of describers and a plurality of labels. When there is not or rarely positive example whose desired label is turned to be a desired value in the learning data storage part 131, a control part 150 generates a temporary positive example by rewriting the value of the desired label with the value of another similar label. An active learning part 140 learns a rule by using the temporary positive example and a negative example, and predicts the positive example likeliness of each candidate data by applying the learned rule to the group of candidate data stored in a candidate data storage part 133 whose desired label is unknown, and selects data to be learnt the next based on the prediction result, and outputs the data form an input/output device 110. Afterwards, the data whose desired label's actual values are inputted from an input/output device 110 are removed from the group of the candidate data, and added to the group of the learning data. COPYRIGHT: (C)2007,JPO&INPIT
|