In the new energy vehicle industry, precise identification of core technologies is the key to promoting innovation and maintaining market competitiveness. In this article, a comprehensive approach combining information weight method and SAO-BERTopic topic model is proposed to extract and analyze core technologies from large-scale patent data. Through in-depth analysis of Incopat patent database, we use the information weight method to select high-quality core patents from four dimensions: technological, strategic, law and market value. These selected patents form the basis of the research data, which is then applied to the SAO-BERTopic model, which combines the advanced semantic understanding capabilities of BERTopic with the fine-structured characteristics of SAO analysis, greatly improving the efficiency and accuracy of identifying complex technical topics. This innovation of this research is not only applicable to the analysis of the technological development of the new energy vehicle industry, but also can provide valuable reference for other high-tech industries such as biomedicine, renewable energy and information technology. These fields also need to identify core technologies from a large number of patents. SAO-BERTopic's structured analysis framework can help these industries to insight into technology development trends, identify innovation opportunities, and provide data-driven decision support for enterprises, research institutions and governments, thus playing an important role in technology planning and market commercialization.
the rapid development of the new energy automobile industry has become an important symbol of technological innovation and industrial transformation[1].
field[2,3]. However, in the face of the strategic layout and potential containment of traditional automobile powers such as the United States, Europe, Japan and South Korea in terms of technology and industrial chain, the sustainable growth and competitiveness of China's new energy automobile industry presents new challenges [4,5,6,7].
and innovation, patents and scientific papers are indispensable resources. However, due to the large scale of these literatures, traditional manual analysis methods are difficult to cope with, and expert analysis is subjective. Therefore, based on the large amount of technical information contained in patents [8], many scholars have devoted themselves to using data mining methods in recent years [9]. Among them, the topic model is an approach that can automatically extract key topics from documents, revealing technology trends and areas of innovation. Latent Dirichlet Allocation (LDA) is a word frequency-based models that may struggle to capture semantic complexity [10,11,12,13]. Latent Semantic analysis (LSA) has limitations in handling word sense diversity and polysemy, which can lead to information loss when analyzing specialized technical documents [14,15]. Correlated Topic models (CTM) high computational complexity limits its rapid application on large document collections [11,16] .
In view of the limitations of existing topic models in the identification of core technologies in the new energy vehicle industry, we propose an innovative topic model— SAO-BERTopic. The model combines SAO analysis and BERTopic, an advanced natural language processing technology, to improve the accuracy and efficiency of technology recognition.
The main innovations of this article are: 1. By combining SAO analysis and BERTopic topic model, we construct SAOBERTopic method, which effectively solves the problem of identifying deep-level technology trends and key innovation points in the new energy vehicle industry.
2. By using the combination of the information weight method and the SAOBERTopic model, we realize the selection of highly innovative and market-valued core patents from large-scale data in the patent analysis of the new energy vehicle industry for the first time, and improve the efficiency and accuracy of technology identification.
We adopt an innovative hybrid approach, utilizing a phased scientific process to identify the core technological topics within the new energy vehicle industry. This process is divided into six detailed steps (as shown in Figure 1), designed to ensure the efficient and systematic extraction, analysis, and determination of core technologies from a broad range of patent data. After selecting patents related to core technologies using the information weight method from a large pool of patents, the SAO-BERTopic model is applied for an in-depth analysis of the selected patents. This model identifies the core technologies' SAO triplets and defines these triplets as specific topics.
Step One: Patent Selection — The first step is to use Incopat patent database to download patents, and use the information weight method to establish indicators in four dimensions of patent technical, strategic, legal and market value; Step Two: Textual Vector Representation — This step is text preprocessing and then vectorization via BERT(Bidirectional encoder representation from the transformer) model; Step Three: Dimensionality Reduction and Clustering — In this step, UMAP(Uniform manifold Approximation and projection) algorithm is used to reduce the dimensionality of the text vector, and then HDBSCAN(hierarchical densi-based applied spatial clustering with noise) algorithm is used to cluster the topic; Step Four: SAO Triplet Extraction — Using the C-TF-IDF(class-based word frequencyinverse document frequency) algorithm, SAO triples are extracted from the clusters and the topic representation is refined using the MMR(Maximum marginal correlation) algorithm to balance correlation and diversity; Step Five: Result Validation — The Calinski-Harabasz index and Davies-Bouldin index were used to compare the LDA, BERTopic and SAO-LDA topic models to evaluate the clustering effectiveness. In addition, dimensionality reduction visualization via UMAP provides an intuitive comparison of cluster distributions; Step Six: Strategic Recommendations — Based on the results, recommendations are made from the corporate, industry and governance perspectives.
eMnwCeIaenriosfgogshrrtyimevmaaeettiteouhpnoctdahotnmeonoltobPPPPsgaaaaittttliyeeeeennnnnttttpitltsmneetharcgadahrtaetklneeueigvnnctsiaacvlettluavrwvelauayleluuee prTCepeorxorcteeVstseienccghBtontEeroxRilztoSPTLTagtOeootmkySpieomnwtpainaogztaargizdtwitiansoetgfinionitlnthtering VReGcDeCetdoniHlmurueDscrsettaunBietossrSiiinnsnoCgugnoAfsTaUiNlTnoiMtpgeyxiActP
PatenSttSecpr1eening Vector ReSpTtreeexpst2entation of ClusterDRoeScputremepse3ennttsation of
Topic 0 is "Energy Management and Power Transmission Technology in Hybrid Electric Vehicles". Topic 1 is "Electric Vehicle Charging Systems and Energy Management", where the SAO triples contained in topic 1 focus on describing the functionalities of receiving, managing, and providing energy in charging panels and systems. Topic 2 is "Battery System Integration and Energy Efficiency Management", where the SAO triples collectively depict various aspects of energy storage system design, management, and application, including modular composition, energy control and management, system power supply, and integration with electric vehicles. Topic 3 is "Electric Vehicle Battery Pack Configuration and Structural Design", covering various aspects of the physical configuration, structural design of battery packs, and how these designs impact battery pack performance. Topic 4 is "Cathode Materials and Composition for Secondary Batteries", focusing on the composition and design of cathode materials for batteries, especially secondary batteries, which directly impact battery performance aspects like capacity, energy density, charge/ discharge speed, and lifespan.
This research successfully applied the information weight method to screen highquality core patents from a large-scale patent dataset, and employed the SAO-BERTopic model to identify core technological topics from the filtered data. Our empirical analysis results highlight the superior clustering performance of SAO-BERTopic in identifying technological topics in the new energy vehicle domain compared to traditional models. The approach presented in this research, through its efficient processing of technical terms and complex concepts and its precise screening of patent data, demonstrates the ability to identify core technologies and drive innovation across a wide range of fields. Therefore, this research not only has a direct contribution to the technological progress of the new energy vehicle industry, but also provides a new perspective and tool for technology identification and innovation management in various fields, which is helpful to promote the technological development and innovation strategy formulation in a wider range of fields. [9] N. Su, Z. Tan, Review and vision for the future of the research on technology [1] Y. Li, The impact of economic systems opportunity analysis methods, Inf. Stud. and financial systems on new energy Theory Appl. 43 (2020) 179–186. vehicle industry, Adv. Econ. Manag. [10] D. M. Blei, A. Y. Ng, M. I. Jordan, Latent Polit. Sci. 33 (2023) 162-170. dirichlet allocation, J. Mach. Learn. Res. doi:10.54254/2754-1169/33/20231622. 3 (2003) 993-1022. [2] Z. Liu, H. Hao, X. Cheng, F. Zhao, [11] T. L. Griffiths, M. Steyvers, Finding Critical issues of energy efficient and scientific topics, Proc. Natl. Acad. Sci. U. new energy vehicles development in S. A. 101 (2004) 5228-5235. china, Energy Policy 115 (2018) 92-97. doi:10.1073/PNAS.0307752101. doi:10.1016/J.ENPOL.2018.01.006. [12] J. Lafferty, D. Blei, Topic models, in: A. [3] P. Yu, J. Zhang, D. Yang, X. Lin, T. Xu, N. Srivastava and M. Sahami (Eds.) Text The evolution of China's new energy mining, Chapman and Hall/CRC, New vehicle industry from the perspective of York, NY, 2009, pp. 71-93. a technology-market-policy framework, doi:10.1201/9781420059458.CH4. Sustainability 11 (2019) 1711. [13] M. Hoffman, F. Bach, D. Blei, Online doi:10.3390/SU11061711. learning for latent dirichlet allocation, in: [4] M. Kendall, Fuel cell development for 24th Annual Conference on Neural new energy vehicles (NEVs) and clean Information Processing Systems 2010, air in china, Prog. Nat. Sci. Mater. Int. 28 NeurIPS, New York, NY, 2010, pp. 856– (2018) 113-120. 864.
doi:10.1016/J.PNSC.2018.03.001. [14] S. Deerwester, S. T. Dumais, G. W. [5] T. Yang, C. Xing, X. Li, Evaluation and Furnas, T. K. Landauer, R. Harshman, analysis of new-energy vehicle industry Indexing by latent semantic analysis, J. policies in the context of technical Am. Soc. Inf. Sci. 41 (1990) 391-407. innovation in china, J. Clean. Prod. 281 doi:10.1002/(SICI)1097(2021) 125126. 4571(199009)41:6<391::AIDdoi:10.1016/J.JCLEPRO.2020.125126. ASI1>3.0.CO;2-9. [6] X. L. Xu, H. H. Chen, Exploring the [15] T. K. Landauer, P. W. Foltz, D. Laham, innovation efficiency of new energy An introduction to latent semantic vehicle enterprises in china, Clean analysis, Discourse Process 25 (1998) Technol. Environ. Policy 22 (2020) 1671- 259-284. doi:10.1080/01638539809545028. 1685. doi:10.1007/S10098-020-01908-W. [16] J. Lafferty, D. Blei, Correlated topic [7] S. Chen, Y. Feng, C. Lin, Z. Liao, X. Mei, models, in: Proceedings of the 18th Research on the technology innovation International Conference on Neural efficiency of China's listed new energy Information Processing Systems, MIT vehicle enterprises, Math. Probl. Eng. Press, Cambridge, MA, 2005, pp. 147– 2021 (2021) 6613602. 154. doi:10.5555/2976248.2976267. doi:10.1155/2021/6613602. [8] C. Lee, A review of data analytics in technological forecasting, Technol.
Forecast. Soc. Change 166 (2021) 120646. doi:10.1016/J.TECHFORE.2021.120646.