MODIFICATION RELATION ANALYSIS PARAMETER LEARNING DEVICE, MODIFICATION RELATION ANALYSIS DEVICE, METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To enable accurate modification analysis.SOLUTION: A learning characteristic extraction unit 24 extracts a characteristic vector composed of the amount of characteristics on the basis of a leaning document, regarding a pair of character string units corresponding to a pair of nodes coupled by an edge and a combination of relation labels given to the edge for each of edges of a discourse structure tree on the basis of a modification structure converted by a rhetoric structure tree conversion unit 22. A parameter learning unit 26 learns a parameter vector optimized on the basis of a score calculated based on the characteristic vector for each of the edges extracted by the learning characteristic extraction unit 24 and a parameter vector composed of weights of the amount of the characteristics of the characteristic vector, and a score calculated on the basis of the characteristic vector extracted for each of the edges of the discourse structure tree on the basis of any dependence structure generated from a series of the character string units of the learning document and the parameter vector.