Measuring Similarity of Story Development for Picture Book Search

The purpose of this research is to build a system for searching picture books based on the similarity of story lines. Current systems search for books by referring to "outline" sources that relate to the book, such as bibliographic information; content-based searches are difficult to perform. As a consequence, existing book search systems do not respond to user requests like, "Please show me a picture book that gives me a similar postreading feeling as this picture book." To handle such requests, this paper focuses on the sentiment polarity of words: our proposed method enables similarity search of picture books by measuring transitions in the sentiment polarity of words that appear in a picture book's text. In this method, positive and negative words appearing in the text are extracted, and the similarity between picture books is calculated using Bhattacharyya coefficients. Our experiment reveals that the proposed method finds books similar to the query book in terms of story lines.