[Objective/Significance] With the rapid development of China's new energy automobile industry, enterprises urgently need to accurately capture user demand characteristics to drive continuous optimization and innovation of products and services. However, given the rapid updates and frequent hotspot shifts of Weibo topics, traditional demand analysis methods exhibit significant limitations in topic recognition comprehensiveness, demand hierarchy relevance mining, and dynamic evolution tracking. [Method/Process] By analysing the interaction data under the topic of new energy vehicle “BYD Han” in Sina Weibo, firstly the study employs the BERTopic model for topic modeling to identify and analyze hotspot topics. Secondly, it combines Maslow's Hierarchy of Needs Theory to analyze topic relevance, revealing the distribution characteristics of demand hierarchies. Finally, the Dynamic Topic Model (DTM) is used for time-series modeling of topics to uncover the evolution trends of user demands. [Results/Conclusions] The results show that hotspot analysis reveals user discussions mainly focus on new car releases, Chinese-style design, and product sales. Relevance analysis combined with Maslow's theory demonstrates that highly relevant topics are mostly concentrated in adjacent or the same hierarchy, verifying the hierarchical characteristics of user demands. Theme evolution analysis indicates that industry hotspot events significantly guide public attention, intensified market competition drives rationalization of consumption decisions, and brand marketing strategies profoundly influence user behavior. Due to its high consistency, the BERTopic model can identify and track changes in user demand more accurately than the LDA model.
Yue Fang, Li Qi, Pan Chonglin, Dai Wenhui, Lu Wen.
Research on User Demand Characteristics and Evolution of New Energy Vehicles: Topic Mining Dased on the BERTopic Model[J].
Science Focus, 2025, 20(5): 58-74 DOI:10.15978/j.cnki.1673-5668.20250402
主题模型是从大量非结构化文本数据中提取出具有实质意义的主题,进而理解用户讨论的核心内容及发展趋势[11]。针对新能源汽车,用户的讨论内容不仅涉及技术,还涉及续航里程、充电便利性、价格、智能化配置等更广泛的内容。此外,新能源汽车的技术和车型更新速度快,新技术、新设计不断涌现,用户关注点不断变化。因此,为了对新能源汽车话题下的交互数据进行深入分析,把握热点主题、主题间的内在联系和动态演化,可利用主题模型进行相关分析。而目前常用的LDA(latent dirichlet allocation)、word2vec(word to vector)等模型在处理结构化主题和静态文本方面具有一定优势,但在应对语义变化快、文本长度短、话题更新快等场景下存在一定局限。LDA处理短文本或稀疏数据时效果不佳[12],word2vec的静态词向量难以适应语义的动态变化[13]。为此,需要寻找新的模型。
(ZhouJ. Intelligent manufacturing:main direction of“made in China 2025”[J]. China Mechanical Engineering, 2015, 26(17): 2273-2284.)
[3]
XuY Y, ShanX Y, GuoM C, et al. Design and application of experience management tools from the perspective of customer perceived value: a study on the electric vehicle market[J]. World Electric Vehicle Journal, 2024, 15(8): 378.
[4]
TheodorakopoulosL, TheodoropoulouA. Leveraging big data analytics for understanding consumer behavior in digital marketing: a systematic review[J]. Human Behavior and Emerging Technologies, 2024, 2024(1): 3641502.
(YanJ Y, MaS C. Research on network microblogging lyrics topic recognition and tracking technology based on clustering[J]. Journal of Chongqing University of Technology (Natural Science), 2019, 33(9): 176-181.)
(MiL Y, YuX Y, YangJ. Research on influencing factors of knowledge consumers'purchase behavior of new energy vehicles:based on the exploration of grounded theory[J]. Enterprise Economy, 2018, 37(4): 19-26.)
(WangK Q, GaoZ J, QiaoY N, et al. User requirements identification and trend of evolution mining in online reviews[J]. Mechanical Science and Technology for Aerospace Engineering, 2023, 42(7): 1070-1080.)
(WangZ Y. Driving factors and mechanism of online reverse social emotion in emergencies: a csQCA analysis based on 30 cases[J]. Journal of Information Resources Management, 2024, 14(5): 116-131.)
(ZhaoF F, WangY Q, KongQ C, et al. Application framework and empirical study of web media content analysis on public policy:new energy vehicle policy[J]. Information Science, 2020, 38(4): 140-146.)
(YeG H, LiS Y, XieS, et al. Multidimensional analysis of open-source technological intelligence in the new energy vehicle industry[J]. Information Science, 2024, 42(2): 155-164.)
(ShiW, QuY J, WangX S. Supervised topic modeling and application research on product recall considering textual structural features[J]. Chinese Journal of Management Science, 2024: 1-19 [2025-07-05]. https://link.cnki.net/doi/10.16381/j.cnki.issn1003-207x.2023.1637.)
(WangL, LiS X, ChenY. Public opinion evolution of COVID-19 vaccine based on topic-emotion fusion analysis[J]. Information Science, 2024, 42(8): 65-75.)
(TengF, ZhangQ, QuJ S, et al. Identifying key and core technologies based on patent competitiveness index and Doc-LDA topic model: case study of E-vehicles[J]. Data Analysis and Knowledge Discovery, 2024, 8(11): 33-46.)
(MaoY Y, LuW. The bilateral analysis framework of patented technology and industry based on IPC association[J]. Information Science, 2022, 40(4): 33-39.)
[29]
WangZ J, ZhouR F, WangY J. DTM-based analysis of hot topics and evolution of China's energy policy[J]. Sustainability, 2024, 16(19): 8293.
[30]
WangS S, MaiS. Study on the theme evolution and synergy assessment of China's new energy vehicle policy texts[J]. Sustainability, 2024, 16(17): 7260.
[31]
ZuhrohS, RiniG P. Product innovation capability and distinctive value positioning drivers for marketing performance: a service-dominant logic perspective[J/OL]. International Journal of Innovation Science, 2024[2025-07-05]. https://www.emerald.com/ijis/article-abstract/doi/10.1108/IJIS-04-2024-0100/1250561.
(KangX, MaoQ R. The influence-consequence mechanism of iterative innovation: a grounded theory analysis based on we chat from tencent[J/OL]. Science & Technology Progress and Policy, 2025: 1-11[2025-07-05]. https://kns.cnki.net/kcms/detail/42.1224.G3.20250523.1643.010.html.)
(ZhangJ J, HuangR J, ChengJ. The construction and implementation path of emotional implantation model for digital textbooks[J/OL]. Modern Distance Education, 2025: 1-16[2025-07-05]. https://link.cnki.net/doi/10.13927/j.cnki.yuan.20250630.001.)
(ZhaoD Y, TangZ J, SunF X. Weak demand signal identification based on a dynamic topic model: taking BYD's new energy vehicles as an example[J]. Journal of Intelligence, 2024, 43(10): 81-89.)
(WangX H, FengX T. Methodological implications of the perspective of multi-subject construction for surveys[J]. Nanjing Journal of Social Sciences, 2017(5): 46-52.)
[42]
WangX G, LüT, FanL. New energy vehicle consumer demand mining research based on fusion topic model: a case in China[J]. Sustainability, 2022, 14(6): 3316.
[43]
HerfortJ D, TamborgA L, MeierF, et al. Twenty years of research on technology in mathematics education at CERME: a literature review based on a data science approach[J]. Educational Studies in Mathematics, 2023, 112(2): 309-336.
(GaoC L, JiangL Y, DongT Y. Thematic evolution of health information needs in the elderly based on BERTopic model[J]. Information Science, 2024, 42(4): 111-118.)
[46]
SunB, JuZ F. Research on the promotion of new energy vehicles based on multi-source heterogeneous data: consumer and manufacturer perspectives[J]. Environmental Science and Pollution Research, 2023, 30(11): 28863-28873.
[47]
LiZ Y, ZouZ F. Punctuation and lexicon aid representation: a hybrid model for short text sentiment analysis on social media platform[J]. Journal of King Saud University - Computer and Information Sciences, 2024, 36(3): 102010.
[48]
KimH, KangH, ChoiH, et al. Human-building interaction for indoor environmental control: evolution of technology and future prospects[J]. Automation in Construction, 2023, 152: 104938.
(YaoJ, MengX L. Research on hot topic of E-commerce microblog based on improved LDA algorithm[J]. Journal of Chongqing University of Technology (Natural Science), 2019, 33(12): 184-188.)
(FanZ T, NingJ. Development priority: government role orientation and temporal and spatial characteristics in local artificial intelligence governance policy discourse[J]. E-Government, 2025(6): 47-62.)
[53]
GrootendorstM. BERTopic: neural topic modeling with a class-based TF-IDF procedure[EB/OL]. [2024-09-07]. https://arxiv.org/abs/2203.05794.
[54]
KhodeirN, ElghannamF. Efficient topic identification for urgent MOOC forum posts using BERTopic and traditional topic modeling techniques[J]. Education and Information Technologies, 2025, 30(5): 5501-5527.
[55]
EggerR, YuJ. A topic modeling comparison between LDA, NMF, Top2Vec, and BERTopic to demystify Twitter posts[J]. Frontiers in Sociology, 2022, 7: 886498.
(LiuY, LiuZ X, JinH, et al. User hierarchical need and motivation analysis ased on BERTopic model: taking Douyin platform as an example[J]. Journal of Intelligence, 2023, 42(12): 159-167.)
[58]
LuW, HangnaZ, TangY. BTVis: an interactive hierarchical topic modelling visual analysis system based on BERTopic[J/OL]. Journal of Computer-aided Design & Computer Graphics, 2025[2025-04-01]. https://www.jcad.cn/en/article/doi/10.3724/SP.J.1089.2024-00082.
[59]
WangZ Y, ChenJ, ChenJ P, et al. Identifying interdisciplinary topics and their evolution based on BERTopic[J]. Scientometrics, 2024, 129(11): 7359-7384.
(MaZ H, HouJ C, XieC X, et al. Research on spectral clustering algorithm based on user characteristics[J]. Information Technology and Informatization, 2023(10): 104-108.)
(LongY X, WangX G, ZhouZ W, et al. Research on text time window partition based on LDA model[J]. Science Focus, 2024, 19(2): 34-45.)
[64]
LiuH C, LiuR, GuX Z, et al. From total quality management to Quality 4.0: a systematic literature review and future research agenda[J]. Frontiers of Engineering Management, 2023, 10(2): 191-205.
(HeC C, HuangQ, LiX R, et al. Trending topics on metaverse: a microblog text analysis with BERT and DTM[J]. Data Analysis and Knowledge Discovery, 2023, 7(9): 25-38.)
MengY, ZhangY Y, HuangJ X, et al. Topic discovery via latent space clustering of pretrained language model representations[C]// Proceedings of the ACM Web Conference 2022.New York, NY, USA: Association for Computing Machinery, 2022: 3143-3152.
(HuK X, LiX, WangL T. Detection of derived public opinion of cyber violence incidents based on BERTopic model[J]. Journal of Intelligence, 2024, 43(7): 146-153.)
(WangL L, LiuY L. Hot technology topics and evolution analysis in the field of computer vision based on BERTopic[J]. Science Focus, 2024, 19(2): 46-60.)
[74]
ZhangT T, YuanB. Density-based multiscale analysis for clustering in strong noise settings with varying densities[J]. IEEE Access, 2018, 6: 25861-25873.
[75]
Blanco-PortalsJ, PeiróF, EstradéS. Strategies for EELS data analysis. introducing UMAP and HDBSCAN for dimensionality reduction and clustering[J]. Microscopy and Microanalysis, 2022, 28(1): 109-122.
(YaoY Z, LiJ L. Feature weight analysis and improvement of TF-IDF based on category information[J]. Computer Systems & Applications, 2021, 30(9): 237-241.)
(XuD D, WuS B. An improved TF-IDF feature selection based on categorical description[J]. New Technology of Library and Information Service, 2015(3): 39-48.)
(SuW, YuS, ZhuoC Y. Analysis of the evolution and identification of topics in China's data element policies[J]. Library and Information Service, 2024, 68(16): 90-103.)
(SunJ Y. Research on incentive mechanism of civil servants' integrity:based on maslow's hierarchy of needs theory[J]. Leadership Science, 2017(5): 56-58.)
(PengX Y, YinZ Y, TaoT, et al. Analysis of the application of office seating demand evaluation system based on PSO-FAHP model[J]. China Forest Products Industry, 2023, 60(2): 46-51.)
(HuangJ H, JinY, ZhangJ. How can enterprise microblogging improve consumer loyalty: an empirical study based on social identity theory[J]. Nankai Business Review, 2016, 19(4): 159-168.)
(LiuX L, YangC Y. Coal mine safety risk topic mining based on BERTopic[J/OL]. Journal of Chongqing Technology and Business University (Natural Science Edition), 2025: 1-9[2025-07-05]. https://kns.cnki.net/kcms/detail/50.1155.N.20250226.1034.002.html.)