Abstract: In a semantic representation generation method, syntax data is generated by a superficial analysis on text data described in a natural language. A concept tag is provided to each morpheme based on the syntax data with reference to a CT system table in which concept information hierarchically and ambiguously representing a meaning of the morpheme for a part of speech. Provided is a first semantic tag indicating semantic information representing a semantic relation between a first phrase/sequence of phrases and a second phrase/sequence of phrases to a first pair made up of the first phrase/sequence of phrases corresponding to a predicate and the second phrase/sequence of phrases having a modification relation with the predicate based on the syntax data. Semantic representation data is generated based on the concept tag provided to each morpheme and the first semantic tag provided to the first pair.

Abstract: Training data in which tag information is assigned to some document files extracted from a plurality of document files to be retrieved is acquired by a training data acquirer. A tag estimation model for estimating tag information to be assigned to a document file is constructed by a constructor by applying the acquired training data to a Transformer machine learning model on which learning has been carried out in advance using a corpus. The tag information is assigned to each of the plurality of document files to be retrieved by an assigner using the constructed tag estimation model.

Abstract: A vector acquisition method includes: inputting, into a learned model, at least one piece of text including at least two of a plurality of words obtained by dividing a compound, the compound being a word divisible into the plurality of words; outputting, from the learned model, an adjusted vector corresponding to at least one of the words obtained by dividing the compound in the input piece of text; and acquiring a compound vector corresponding to the compound using the adjusted vector output from the learned model. Classification accuracy of vectors corresponding to words can thereby be enhanced.

Abstract: Learning data representing the relationship between explanatory variables and objective variables is acquired by an acquirer. In the learning data acquired by the acquirer, a hierarchical relationship among a plurality of items included in the objective variable is determined by a hierarchy determiner. A construction algorithm to be executed among a plurality of construction algorithms for construction of a learning model is determined by an algorithm determiner based on the hierarchical relationship determined by the hierarchy determiner. A first learning model is constructed by execution of the construction algorithm determined by the algorithm determiner by a learner.

Abstract: A parameter update apparatus according to the present invention includes: an input unit configured to receive input of teaching data; and an update unit configured to update a parameter for assigning at least one estimation label corresponding to each of a plurality of data items by performing multi-task learning by using a neural network for the plurality of data items of the input teaching data. The update unit updates the parameter so that a sum of errors between the assigned estimation label and a corresponding true label in the teaching data in the plurality of data items has a minimum value. Therefore, the plurality of data items constituting a hierarchical structure can be classified while preventing deterioration of classification accuracy.