Multidimensional Danmaku Analytics via a BERT-SVM Fusion Model

Ya Lin, Xudong Zhang, Guangbin Peng, Xiang He, Yuanxia Deng
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:379-394, 2025.

Abstract

Danmaku (bullet comments), characterized by real-time interactivity, high concurrency, and textual fragmentation, present unique challenges for semantic analysis in film audience feedback research. To address the limitations of conventional methods in processing sparse short texts and imbalanced data distributions, this study proposes a BERT-SVM fusion model integrating BERT-based semantic representation with SVM classification, supplemented by SMOTE oversampling. Validated on 450,000 Danmaku comments from The Wandering Eart series, the framework achieves a sentiment classification accuracy of 92.6%. Furthermore, a multidimensional analysis pipeline is implemented, combining BERT embedding compression, KMeans clustering, and LDA topic modeling to systematically identify audience discussion themes. Experimental results demonstrate that The Wandering Earth 2 not only elicits a higher proportion of positive sentiment than its predecessor but also shifts thematic focus toward advanced sci-fi elements such as digital life and lunar crisis resolution. This work establishes an efficient analytical framework for large-scale Danmaku data, offering actionable insights to enhance narrative design and audience engagement strategies in the film industry.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-lin25a, title = {Multidimensional Danmaku Analytics via a BERT-SVM Fusion Model}, author = {Lin, Ya and Zhang, Xudong and Peng, Guangbin and He, Xiang and Deng, Yuanxia}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {379--394}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/lin25a/lin25a.pdf}, url = {https://proceedings.mlr.press/v278/lin25a.html}, abstract = {Danmaku (bullet comments), characterized by real-time interactivity, high concurrency, and textual fragmentation, present unique challenges for semantic analysis in film audience feedback research. To address the limitations of conventional methods in processing sparse short texts and imbalanced data distributions, this study proposes a BERT-SVM fusion model integrating BERT-based semantic representation with SVM classification, supplemented by SMOTE oversampling. Validated on 450,000 Danmaku comments from The Wandering Eart series, the framework achieves a sentiment classification accuracy of 92.6%. Furthermore, a multidimensional analysis pipeline is implemented, combining BERT embedding compression, KMeans clustering, and LDA topic modeling to systematically identify audience discussion themes. Experimental results demonstrate that The Wandering Earth 2 not only elicits a higher proportion of positive sentiment than its predecessor but also shifts thematic focus toward advanced sci-fi elements such as digital life and lunar crisis resolution. This work establishes an efficient analytical framework for large-scale Danmaku data, offering actionable insights to enhance narrative design and audience engagement strategies in the film industry.} }
Endnote
%0 Conference Paper %T Multidimensional Danmaku Analytics via a BERT-SVM Fusion Model %A Ya Lin %A Xudong Zhang %A Guangbin Peng %A Xiang He %A Yuanxia Deng %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-lin25a %I PMLR %P 379--394 %U https://proceedings.mlr.press/v278/lin25a.html %V 278 %X Danmaku (bullet comments), characterized by real-time interactivity, high concurrency, and textual fragmentation, present unique challenges for semantic analysis in film audience feedback research. To address the limitations of conventional methods in processing sparse short texts and imbalanced data distributions, this study proposes a BERT-SVM fusion model integrating BERT-based semantic representation with SVM classification, supplemented by SMOTE oversampling. Validated on 450,000 Danmaku comments from The Wandering Eart series, the framework achieves a sentiment classification accuracy of 92.6%. Furthermore, a multidimensional analysis pipeline is implemented, combining BERT embedding compression, KMeans clustering, and LDA topic modeling to systematically identify audience discussion themes. Experimental results demonstrate that The Wandering Earth 2 not only elicits a higher proportion of positive sentiment than its predecessor but also shifts thematic focus toward advanced sci-fi elements such as digital life and lunar crisis resolution. This work establishes an efficient analytical framework for large-scale Danmaku data, offering actionable insights to enhance narrative design and audience engagement strategies in the film industry.
APA
Lin, Y., Zhang, X., Peng, G., He, X. & Deng, Y.. (2025). Multidimensional Danmaku Analytics via a BERT-SVM Fusion Model. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:379-394 Available from https://proceedings.mlr.press/v278/lin25a.html.

Related Material