Abstract
Comics are entertainment contents that conversa- tions of characters drive stories. Since people can obtain new knowledge from conversations in real lives, the conversations in comics could be useful to acquire new knowledge. The conversations in comics contain various utterance sentences of characters. Some of the utterance sentences contain knowledge, and the others not. We define the utterance sentence as a knowledge explanation sentence. If the knowledge explanation sentences are extracted automatically, those can be used for knowledge acquisition with comics and comics recommendation.
In this paper, we propose an extraction method of the knowledge explanation sentences from characters’ conversations in comics. We assume that the knowledge explanation sentences include content words and expressions that are special for knowl- edge explanation. The proposed method represents a sentence as a combination vector of an averaged word vector and an expression vector. The averaged word vector is an average of each content word’s embedding. The expression vector has factors of expression meanings that are used for knowledge explanation. The combination vectors are trained by Support Vector Machine to obtain a classifier. The classification outputs a set of knowledge explanation sentences that will be an extraction result.
We tested the proposed method by comparing other methods with 4,611 sentences from five comics. The proposed method obtained 0.80 accuracy, 0.77 precision, and 0.88 recall that were the highest among the prepared methods.
Information
タイトル：A Method to Extract Knowledge Explanation Sentences from Conversations in Comics with Combination of Contents and Expressions
著者：Yoko Nishihara, Kohei Matsuoka, and Ryosuke Yamanishi
概要：Comics are entertainment contents that conversa- tions of characters drive stories. Since people can obtain new knowledge from conversations in real lives, the conversations in comics could be useful to acquire new knowledge. The conversations in comics contain various utterance sentences of characters. Some of the utterance sentences contain knowledge, and the others not. We define the utterance sentence as a knowledge explanation sentence. If the knowledge explanation sentences are extracted automatically, those can be used for knowledge acquisition with comics and comics recommendation.
In this paper, we propose an extraction method of the knowledge explanation sentences from characters’ conversations in comics. We assume that the knowledge explanation sentences include content words and expressions that are special for knowl- edge explanation. The proposed method represents a sentence as a combination vector of an averaged word vector and an expression vector. The averaged word vector is an average of each content word’s embedding. The expression vector has factors of expression meanings that are used for knowledge explanation. The combination vectors are trained by Support Vector Machine to obtain a classifier. The classification outputs a set of knowledge explanation sentences that will be an extraction result.
We tested the proposed method by comparing other methods with 4,611 sentences from five comics. The proposed method obtained 0.80 accuracy, 0.77 precision, and 0.88 recall that were the highest among the prepared methods.
書籍情報：the 2020 International Conference on Technologies and Applications of Artificial Intelligence, pp.12-16
発表種別：国際会議論文
発表日：2020年 12月3日