摘要 |
PROBLEM TO BE SOLVED: To provide a topic model learning device that outputs a topic sharing model for achieving flexible data mining from one set of two pieces of interaction text information.SOLUTION: A topic sharing model initialization unit outputs a plurality of parameters of each latent variable series of a series and b series and class variable series and others as initial parameters, in which a latent variable and a class variable being a variable for distinguishing between speaker-dependent and topic-dependent are assigned to a word of each text of the a series and b series of an interaction text. A topic sharing model learning unit sequentially updates the initial parameters with a Markov chain Monte-Carlo method by using the initial parameters and hyper parameters inputted from the outside as input, and outputs the parameters. A topic sharing model information output unit outputs topic sharing model information by using all parameters updated by using the Markov chain Monte-Carlo method in the topic sharing model learning unit, and the number of vocabulary types, the number of words and the hyper parameters. |