摘要 |
PROBLEM TO BE SOLVED: To improve structural learning accuracy.SOLUTION: A model learning method comprises: (A) a first step of updating a first score corresponding to each of identities as clues when extracting a correct answer structure from each learning case on the basis of a weight of each learning case, and calculating a model specified by each first score; (B) a second step of identifying a maximum score among second scores allocated by the model to each structure other than the correct answer structure among a plurality of candidate structures derived from the learning case for each learning case, determining a difference between the second score allocated by the model to the correct answer structure of the learning case and the maximum score, and calculating a certainty factor of the model on the basis of an error upper limit defined by the difference in each learning case; (C) a third step of updating the weight of each learning case on the basis of the certainty factor and the difference; and (D) a step of repeating the first to third steps. |