Abstract
In this study, we clarify the differences in emotional characteristics and their temporal changes in the narrativesof popular and general web novels. We collect text data from the Shousetsuka ni Narou website, define the top 300works in the global point ranking as popular works, and randomly select another 300 works as general works. Toperform sentiment analysis at the sentence level, we fine-tune a pre-trained Japanese BERT model using the WRIMEdataset for emotion analysis, classify each sentence into Plutchik ’s eight basic emotions (joy, sadness, anticipation,surprise, anger, fear, disgust, and trust), and extract an 8-dimensional emotion score for each sentence. We dividethe text into fixed-length segments, calculate the average emotion scores as features, construct a classification modelusing Random Forest, and clarify differences in emotional tendencies through SHAP analysis. As a result, we showthat differences in emotional characteristics and their temporal changes in narratives significantly influence a work’spopularity. In the early stages of narratives, anticipation contributes positively to popularity, while surprise and fearcontribute negatively. This suggests that instead of evoking surprise or fear regarding the protagonist’s problems,inducing empathy and a sense of anticipation for their resolution is important for attracting readers emotionally. Inthe middle stages, we find that the development of trust among characters serves as preparation for challenges in thelatter part of the story, playing an important role in the emotional flow of popular works. In the final stages, we observethat popular works maintain emotions related to anticipation and trust as the narrative progresses, while emotions suchas surprise and fear decrease. This suggests that in popular works, strengthening the trust relationship between theprotagonist and companions forms the foundation of the narrative and promotes readers’ emotional immersion.
Translated Abstract
In this study, we clarify the differences in emotional characteristics and their temporal changes in the narrativesof popular and general web novels. We collect text data from the Shousetsuka ni Narou website, define the top 300works in the global point ranking as popular works, and randomly select another 300 works as general works. Toperform sentiment analysis at the sentence level, we fine-tune a pre-trained Japanese BERT model using the WRIMEdataset for emotion analysis, classify each sentence into Plutchik ’s eight basic emotions (joy, sadness, anticipation,surprise, anger, fear, disgust, and trust), and extract an 8-dimensional emotion score for each sentence. We dividethe text into fixed-length segments, calculate the average emotion scores as features, construct a classification modelusing Random Forest, and clarify differences in emotional tendencies through SHAP analysis. As a result, we showthat differences in emotional characteristics and their temporal changes in narratives significantly influence a work’spopularity. In the early stages of narratives, anticipation contributes positively to popularity, while surprise and fearcontribute negatively. This suggests that instead of evoking surprise or fear regarding the protagonist’s problems,inducing empathy and a sense of anticipation for their resolution is important for attracting readers emotionally. Inthe middle stages, we find that the development of trust among characters serves as preparation for challenges in thelatter part of the story, playing an important role in the emotional flow of popular works. In the final stages, we observethat popular works maintain emotions related to anticipation and trust as the narrative progresses, while emotions suchas surprise and fear decrease. This suggests that in popular works, strengthening the trust relationship between theprotagonist and companions forms the foundation of the narrative and promotes readers’ emotional immersion.
情変化の定量的解析
引用文献
- [番庄21] 番庄智也, 片寄晴弘：鬼滅の刃の神回を対象とした感情曲線の分析と検討, 研究報告エンタテインメントコンピューティング(EC), Vol. 2021, No. 12, pp. 1–7 (2021)
- [Chen 20] Chen, Y., Hou, W., Li, S., Wu, C., and Zhang, X.: Endto-end emotion-cause pair extraction with graph convolutional network,in Proceedings of the International Conference on Computational Linguistics, pp. 198–207 (2020)
- [エイデン16] エイデンエレツ, ミシェルジャン=バティースト著,阪本芳久訳：カルチャロミクス: 文化をビッグデータで計測する, 草思社, 東京(2016)
- [Gong 20] Gong, H., Shen, Y., Yu, D., Chen, J., and Yu, D.: Recurrentchunking mechanisms for long-text machine reading comprehension,in Proceedings of the Conference on Empirical Methods in Natural Language Processing, pp. 8311–8324 (2020)
- [花畑19] 花畑圭佑, 青野雅樹：語彙と文脈に着目した文学作品の著者推定, 言語処理学会第25 回年次大会(2019)
- [Huang 19] Huang, Y.-H., Lee, S.-R., Ma, M.-Y., Chen, Y.-H., Yu, Y.-W., and Chen, Y.-S.: EmotionX-IDEA: Emotion BERT–an affectionalmodel for conversation, arXiv preprint arXiv:1908.06264(2019)
- [犬塚25] 犬塚美輪：読めば分かるは当たり前？ ――読解力の認知心理学, No. 480, ちくまプリマー新書(2025)
- [石田10] 石田将吾, 佐藤理史：エッセイコーパスを用いた日本語テキストの著者推定, 情報処理学会研究報告(2010)
- [Jaiswal 23] Jaiswal, A. and Milios, E.: Breaking the token barrier:chunking and convolution for efficient long text classification withBERT, arXiv preprint arXiv:2310.20558 (2023)
- [Kajiwara 21] Kajiwara, T., Chu, C., Takemura, N., Nakashima, Y.,and Nagahara, H.: WRIME: A new dataset for emotional intensityestimation with subjective and objective annotations, in Proceedingsof the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 2095–2104 (2021)
- [Kenton 19] Kenton , J. D. M.-W. C. and Toutanova , L. K.: BERT:Pre-training of deep bidirectional transformers for language understanding,in Proceedings of the North American Chapter of the Associationfor Computational Linguistics: Human Language Technologies,Vol. 1, p. 2 (2019)
- [金21] 金明哲, 中村靖子, 上阪彩香, 土山玄, 孫昊, 劉雪琴, 李広微, 入江さやか：文学と言語コーパスのマイニング, テキストアナリティクス, 第7 巻, 岩波書店(2021)
- [小坂19] 小坂直輝, 小林哲則, 林良彦：隠れた良作を推薦可能なWeb 小説レコメンドシステムの提案, 第23 回インタラクティブ情報アクセスと可視化マイニング研究会(2019)
- [小坂20] 小坂直輝, 小林哲則, 林良彦：隠れた良作の発掘を助けるWeb 小説推薦システムの構成と評価, 第34 回人工知能学会全国大会(2020)
- [Lundberg 17] Lundberg , S. M. and Lee, S.-I.: A unified approachto interpreting model predictions, in Proceedings of the International Conference on Neural Information Processing Systems, Vol. 30, pp.4765–4774 (2017)
- [信国89] 信国佳之：自然言語における長文分割方式, 情報処理学会全国大会(1989)
- [Plutchik 80] Plutchik, R.: A general psychoevolutionary theory ofemotion, in Theories of emotion, pp. 3–33, Elsevier (1980)
- [Sato 25] Sato, T. and Fukazawa, Y.: From planning to prevention:predicting mountain accident risks using pre-climb information, International Journal of Data Science and Analytics, pp. 1–19 (2025)
- [Snyder 05] Snyder, B.: SAVE THE CAT の法則本当に売れる脚本術日本語翻訳版, フィルムアート社, 東京(2005)
- [高田17] 高田叶子, 佐藤哲司：文体の類似度を考慮したオンライン小説推薦手法の提案, データ工学と情報マネジメントに関するフォーラム(2017)
- [Vaswani 17] Vaswani, A.: Attention is all you need, Advances inNeural Information Processing Systems (2017)
- [Vishnubhotla 24] Vishnubhotla, K., Hammond, A., Hirst, G., andMohammad, S.: The emotion dynamics of literary novels, in Findingsof the Association for Computational Linguistics, pp. 2557–2574(2024)
- [Warner 25] Warner, B., Chaffin, A., Clavi´e, B., Weller, O., Hallstr¨om, O., Taghadouini, S., Gallagher, A., Biswas, R., Ladhak, F., Aarsen, T., Cooper, N., Adams, G., Howard, J., and Poli, I.: Smarter,better, faster, longer: a Modern bidirectional encoder for fast,memory efficient, and long context finetuning and inference, arXivpreprint arXiv:25012345 (2025)
- [渡邉25] 渡邉真, 深澤佑介：SHAP による人気Web 小説の物語進行に伴う感情変化の解析, 第39 回人工知能学会全国大会(2025)
- [山崎23] 山崎睦月, 佐藤哲司：小説の特徴量を用いたオンライン小説の検索ワード推薦手法の提案, データ工学と情報マネジメントに関するフォーラム(2023)
- [Yang 20] Yang, L., Zhang, M., Li, C., Bendersky, M., and Najork ,M.: Beyond 512 tokens: siamese multi-depth transformerbasedhierarchical encoder for long-form document matching, in Proceedingsof the ACM International Conference on Information andKnowledge Management, p. 1725–1734 (2020)
- [全国23] 全国出版協会出版科学研究所：出版指標年報2023 年版, 全国出版協会出版科学研究所, 東京(2023), 日本の出版業界に関するデータと分析を収録した年次報告書
- [Zhuo 23] Zhuo, S., Meng, W., Wei, C., and Xiaonan, L.: Researchon emotional classification and literary narrative visualization basedon graph convolutional neural network, in International Conferenceon Image and Graphics, pp. 262–272 (2023)
