This study introduces a deep learning (DL) framework designed to classify YouTube videos into five exercise categories: TABATA, cardiorespiratory endurance, free weight muscular endurance, training aid muscular endurance, and flexibility. These categories were identified by two domain experts in sports science based on video content, tags, and user interaction patterns. Utilizing data from 1284 Traditional Chinese-language YouTube videos and 39,141 comments posted between March 2023 and June 2024, we applied text preparation techniques and sentiment analysis to clean and analyze the dataset. The study incorporated five key features—titles, hashtags, descriptions, comments, and sentiment, and employed Word2Vec with the Continuous Bag of Words (CBOW) model for word embedding, effectively handling short YouTube comments. We evaluated several predictive models, including Support Vector Machine (SVM), Extreme Gradient Boosting (XGBOOST), Text Convolutional Neural Network (TextCNN), Bidirectional Long Short-Term Memory (BiLSTM), CNN-BiLSTM, and BERT, using 10-fold cross-validation and comparing their performance based on accuracy, precision, recall, and F-measure. The results indicate that BERT outperformed the other models, achieving an accuracy of 92.25% (based on 10-fold cross-validation) when all five features were used. Incorporating sentiment analysis enhanced prediction performance across all models, with an average accuracy improvement of 3.85%. Ablation analysis confirmed the significant role of sentiment in improving prediction outcomes. Furthermore, a paired samples t-test revealed statistically significant differences in performance metrics when sentiment was included as a feature. This study demonstrates the effectiveness of combining deep learning and sentiment analysis to predict home-based exercise video categories on YouTube, offering valuable insights into content trends and user engagement in the evolving digital fitness landscape.
HabibNKamalFMunirM. Comparison of the effectiveness of home-based workouts and gym training according to caloric intake. Int Health Rev2021; 1: 13–29.
4.
GuiFTsaiCVajdaA, et al.Workout connections: investigating social interactions in online group exercise classes. Int J Hum Comput Stud2022; 166: 102870.
5.
ParkerKUddinRRidgersND, et al.The use of digital platforms for adults’ and adolescents’ physical activity during the COVID-19 pandemic (our life at home): survey study. J Med Inter Res2021; 23: e23389.
JiaCYeS. Tendencies and characteristics of public fitness behavior in the digital era: an internet ethnography study based on the K community in China. Front Public Health2024; 12: 1414837.
8.
PengXMenhasRDaiJ, et al.The COVID-19 pandemic and overall wellbeing: mediating role of virtual reality fitness for physical-psychological health and physical activity. Psychol Res Behav Manage2022; 15: 1741–1756.
9.
ValerianiFProtanoCDe GiorgiA, et al.Analysing features of home-based workout during COVID-19 pandemic: a systematic review. Public Health2023; 222: 100–114.
10.
VinnikovaALuLWeiJ, et al.The use of smartphone fitness applications: the role of self-efficacy and self-regulation. Int J Environ Res Public Health2020; 17: 7639.
11.
LiuJZhengSXuG, et al.Cross-domain sentiment aware word embeddings for review sentiment analysis. Int J Mach Learn Cybern2021; 12: 343–354.
12.
CronshawS. Web workouts and consumer well-being: the role of digital-physical activity during the UK COVID-19 lockdown. J Consum Aff2021; 56: 449–464.
13.
SullivanANLachmanME. Behavior change with fitness technology in sedentary adults: a review of the evidence for increasing physical activity. Front Public Health2017; 4: 222909.
14.
CunhaAACostaMCPachecoMA. Sentiment analysis of YouTube video comments using deep neural networks. In: Lecture notes in computer science, 2019, pp.561–570. Cham: Springer International Publishing.
15.
MaoYShenYQinG, et al.Predicting the popularity of online videos via deep neural networks, arXiv (Cornell University), 2017. DOI: 10.48550/arXiv.1711.10718.
16.
MiCLiMWulandariAF. Predicting video views of web series based on comment sentiment analysis and improved stacking ensemble model. Electronic Commer Res2022. DOI: 10.1007/s10660-022-09642-9
17.
TianRYinRGanF. Exploring public attitudes toward live-streaming fitness in China: a sentiment and content analysis of China’s social media Weibo. Front Public Health2022; 10: 1027694.
18.
EyipınarCBuyukkalkanFSemizK. Sentiment analysis of YouTube videos comments on sports nutrition. J Stud2021; 2: 27–39.
19.
AlhujailiRFYafoozWMS. Sentiment analysis for YouTube videos with user comments: Review. In: 2021 International conference on artificial intelligence and smart systems (ICAIS), 2021, pp.814–820. IEEE. DOI: 10.1109/ICAIS50930.2021.9396049.
20.
DeciELRyanRM. The “What” and “Why” of goal pursuits: human needs and the self-determination of behavior. Psychol Inq2000; 11: 227–268.
21.
ChenNSunYYanY. Sentiment analysis and research based on two-channel parallel hybrid neural network model with attention mechanism. IET Control Theory Appl2023; 17: 2259–2267.
22.
YouTube Data API. YouTube provides an interface with developers to access video statistics and YouTube channels data, https://developers.google.com/youtube/v3/ (2024, accessed 1 August 2024).
23.
Jieba. Chinese text segmentation: built to be the best Python Chinese word segmentation module, https://github.com/fxsjy/jieba/ (2023, accessed 15 November 2023).
MikolovTChenKCorradoG, et al.Efficient estimation of word representations in vector space. In: Proceedings of international conference on learning representations (ICLR 2013), Scottsdale, Arizona, USA, 2013.
26.
BansalBSrivastavaS. Sentiment classification of online consumer reviews using word vector representations. Procedia Comput Sci2018; 132: 1147–1153.
27.
ChakrabortyKBhattacharyyaSBagR, et al.Comparative sentiment analysis on a set of movie reviews using deep learning approach. In: The international conference on advanced machine learning technologies and applications (AMLTA2018). AMLTA 2018, 2018, Advances in Intelligent Systems and Computing, vol. 723. Cham: Springer.
28.
FigueiredoFAlmeidaJMGonçalvesMA, et al.On the dynamics of social media popularity. ACM Trans Inter Technol2014; 14: 1–23.
29.
NguyenHNguyenM. A deep neural architecture for sentence-level sentiment classification in Twitter social networking, arXiv (Cornell University), 2017. DOI: 10.48550/arxiv.1706.08032.
30.
ChenLLeeCChenM. Exploration of social media for sentiment analysis using deep learning. Soft Comput2019; 24: 8187–8197.
31.
FuFFangSChenW, et al.Sentiment-oriented transformer-based variational autoencoder network for live video commenting. ACM Trans Multimed Comput Commun Appl2023; 20: 1–24.
32.
VapnikV. The nature of statistical learning theory. New York: Springer Verlag, 1995.
33.
ChenTGuestrinC. XGBoost: a scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, 2016, pp.785–794.
GravesASchmidhuberJ. Framewise phoneme classification with bidirectional LSTM and other neural network architectures. Neural Netw2005; 18: 602–610.
36.
DevlinJChangMWLeeK, et al.BERT: pre-training of deep bidirectional transformers for language understanding. In: Proceedings of the 2019 conference of the North American Chapter of the Association for computational linguistics: human language technologies, vol. 1, pp.4171–4186, Minneapolis, Minnesota, June 2019. Association for Computational Linguistics.
37.
Pytorch. Chinese text segmentation: built to be the best Python Chinese word segmentation module, https://pytorch.org/ (2022, accessed 12 January 2022).
38.
SulaimanMSMaskanFADerasitZ, et al.Exploring YouTube comments to understand public sentiment on COVID-19 vaccines through deep learning-based sentiment analysis. Malaysian J Appl Sci2023; 8: 13–27.
39.
KhanGMohaisenMTrierM. The network ROI. Inter Res2019; 30: 631–652.
40.
MedjaniFRutterRNadeauJ. Social media management, objectification and measurement in an emerging market. Int J Bus Emerg Mark2019; 11: 288.
41.
TrunfioMRossiS. Conceptualising and measuring social media engagement: a systematic literature review. Italian J Mark2021; 2021: 267–292.
42.
AnastasiouPTzafilkouKKarapiperisD, et al.YouTube sentiment analysis on healthcare product campaigns: combining lexicons and machine learning models. In: Proceedings of the 14th international conference on information, intelligence, systems and applications (IISA 2023), Volos, Greece, 10–12 July 2023.
43.
BhargavaNRadaidehMIKwonOH, et al.On the impact of language nuances on sentiment analysis with large language models: paraphrasing, sarcasm, and emojis, arXiv preprint arXiv:2504.05603, 2025. DOI: 10.48550/arXiv.2504.05603.
44.
LeeHNguyenMTN. Topic modelling and sentiment analysis on YouTube sustainable fashion comments. J New Media2023; 5: 65–80.
