PROBLEM TO BE SOLVED: To realize efficient learning of an estimation model using an MLN for estimating a relationship between documents.SOLUTION: An instance division part 23 divides an utterance set of learning data into multiple subsets, and a feature amount extraction part 25 extracts an observation predicate showing a feature between utterances and a feature of each utterance, with respect to each subset. A model learning part 28 uses a weight of each logical formula in an estimation model learned to other subsets to estimate, by an MLN, a latent predicate showing the relationship between the utterances, with respect to each subset, and repeatedly learns the weight of each logical formula in the estimation model so that each relationship between the utterances shown by the estimated latent predicate coincides with a previously-given relationship between the utterances.
展开▼