Profiling Wellness Product Consumers in Short-Video Livestreaming Commerce: A Big Data Cluster Analysis

CAO Hongyu, GAN Weixiang

Abstract

Drawing on 186,442 users of wellness-category livestreaming channels on a leading short-video platform, this study merges 90-day transaction records, viewing and interaction logs, and 4.126 million comments into a feature set that spans two modalities, behavior and text. An LDA topic model condenses the comments into three demand-topic intensities (tonic diet, fitness shaping, sleep and mood), the entropy weight method assigns feature weights, and K-means++ performs the clustering. The average silhouette coefficient and the variance ratio criterion both peak at k = 5, yielding five segments: senior tonic (24.1%), white-collar functional (22.0%), highengagement repeat (8.9%), promotion-sensitive (18.0%), and lurker (26.9%). The high-engagement repeat segment delivers 40.8% of gross merchandise volume with 8.9% of users. In a negative binomial regression of purchase counts over a 30-day holdout window, every segment dummy is significant at the 1% level, and purchase intensity in the high-engagement repeat segment is 23.9 times that of the lurker segment. Clustering on transaction features alone blurs the white-collar functional and senior tonic segments and drags the adjusted Rand index down to 0.624. The demand information carried by comment text proves indispensable to segmentation.

Keywords

Livestreaming commerceConsumer profilingCluster analysisLatent Dirichlet allocationNegative binomial regressionWellness products

References

  1. Arthur D, Vassilvitskii S. k-means++: the advantages of careful seeding[C]//Proceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms. Society for Industrial and Applied Mathematics, 2007: 1027-1035.
  2. Blei D M, Ng A Y, Jordan M I. Latent Dirichlet allocation[J]. Journal of Machine Learning Research, 2003, 3: 993-1022.
  3. Caliński T, Harabasz J. A dendrite method for cluster analysis[J]. Communications in Statistics, 1974, 3(1): 1-27.
  4. Cameron A C, Trivedi P K. Regression analysis of count data[M]. 2nd ed. Cambridge: Cambridge University Press, 2013.
  5. CNNIC. The 55th statistical report on China's internet development[R]. Beijing: China Internet Network Information Center, 2025.
  6. CNNIC. The 56th statistical report on China's internet development[R]. Beijing: China Internet Network Information Center, 2025.
  7. Douyin E-commerce, Yicai Business School. White paper on health food consumption trends[R]. 2023.
  8. iiMedia Research. Big data and decision analysis report on the global and Chinese health industry, 2022-2023[R]. 2022.
  9. MacQueen J. Some methods for classification and analysis of multivariate observations[C]// Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability. Berkeley: University of California Press, 1967: 281-297.
  10. Meng L, Liu F J, Chen S Y, et al. Can I arouse you: the influence of information source characteristics of different types of live streamers on consumers' purchase intention[J]. Nankai Business Review, 2020, 23(1): 131-143.
  11. Rousseeuw P J. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis[J]. Journal of Computational and Applied Mathematics, 1987, 20: 53-65.
  12. Smith W R. Product differentiation and market segmentation as alternative marketing strategies[J]. Journal of Marketing, 1956, 21(1): 3-8.
  13. Wedel M, Kamakura W A. Market segmentation: conceptual and methodological foundations[M]. 2nd ed. Boston: Kluwer Academic Publishers, 2000.
  14. Wongkitrungrueng A, Assarut N. The role of live streaming in building consumer trust and engagement with social commerce sellers[J]. Journal of Business Research, 2020, 117: 543-556.
  15. Xie Y, Li C Q, Gao P, et al. The influence and mechanism of social presence on online herd consumption in livestream marketing: a behavioral and neurophysiological perspective[J]. Advances in Psychological Science, 2019, 27(6): 990-1004. DOI:10.3724/SP.J.1042.2019.00990.