摘要 |
PROBLEM TO BE SOLVED: To increase prediction accuracy of labels of contents through active learning.SOLUTION: A correct label acquisition section 14 acquires a correct label assigned by a user to a sample selected from among the plurality of contents. A user certainty factor setting section 16 sets a user certainty factor to the correct label assigned by the user. A label prediction learning section 17 performs label prediction learning. A label prediction/certainty factor calculation section 18 predicts the label of unevaluated contents and calculates a label certainty factor to the predicted label. A user certainty factor prediction section 19 performs user certainty factor prediction learning, and predicts a user certainty factor to the predicted label of the unevaluated contents. A sample selection section 21 selects the next sample from the unevaluated contents. This invention is applicable to a learning device for predicting the label of contents, for instance. |