PROBLEM TO BE SOLVED: To improve quality of an embedded word to be learned.SOLUTION: A tree structure generation unit 24 prepares the tree structure of a text for each plurality of texts. A probability calculation unit 34 calculates probability of the text based on embedment of each node of each layer of the tree structure and a model parameter of a model for predicting embedment of each node. An update section 38 updates embedment of the node of each layer of the tree structure and the model parameter so as to maximize probability of the text. The update section 38 repeats updating until a convergence decision unit 36 satisfies a prescribed convergence decision condition.SELECTED DRAWING: Figure 1
展开▼