Segmentation and Impression Estimation of Novels for Automatic Assignment of BGM

In this paper, we proposed a BGM assignment method to story segments using emotion dictionary. For segmentation of a story, we used information of characters appearing in a story. Changes of appearing characters and characters´ sentiment information are used for segmentation. We prepared syntactic patterns for identification of characters and extraction method of sentiment information. For assignment of BGM to story segments, we extracted sentiment descriptions from story segments and recognized sentiment type of the segments using Emotion Description Dictionary. Detected sentiment information of a segment will be used to generate emotion vector. In the experiments, accuracy of segmentation was not good enough, but we could obtain helpful direction to improve our method, which is to use time and space information. For sentiment information detection, emotion vector model worded well although there are some problems in pattern patching of sentiment expression using Emotion Description Dictionary.