TOPIC MODEL LEARNING METHOD, APPARATUS, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To learn topic models which accords human intuition without increasing a processing load even when a small amount of supervised data is used.SOLUTION: A supervised posterior probability imparting unit 18 imparts supervised posterior probability 42 based on a ratio of topics to key words for each document data 34 including the key words. A posterior probability estimation unit 20a estimates unsupervised posterior probability 44a by using topic model parameter 38 for each document data 34. A weighted posterior probability imparting unit 22 subjects the supervised posterior probability 42 and the unsupervised posterior probability 44a to linear interpolation using weight w(i) acquired from a weight adjustment table 46 to impart weighted posterior probability 48. A posterior probability estimation unit 20b estimates unsupervised posterior probability 44b for each document data 36 which does not include the key words. A parameter update unit 24 updates the topic model parameter 38 on the basis of a predetermined number of topics, the unsupervised posterior probability 44b, and the weighted posterior probability 48.