Method for Automatically Generating Networks of Personal Relationships from Story Summaries

We propose a method of automatically generating networks of personal relationships from program summaries, to enable an understanding of relations between story characters or persons in the programs. The method uses natural language processing to extract representations of the characters or persons from the target text. Then it creates relationships between these story characters or persons from correspondences found through syntax analysis. It determines whether anaphora relations exist between the representations, and it integrates the anaphora relations that refer to characters or persons to generate a network of personal relationships. The results of an evaluation revealed that the networks had Fmeasures of 0.907 for representation, 0.708 for relationship, and 0.800 for anaphora resolution. This shows that it is possible to extract relationships that cannot be extracted by using simply the co-occurrence frequency of personal names and that networks of personal relationships can be generated from short texts such as program summaries. Author