Abstract: Technology opportunity identification using the concept of the gap between science and technology has attracted attention in the academic community. However, there is limited research that considers the exploration of deep semantic features in scientific and technical literature to enhance the interpretability of technology opportunity identification results. This study proposes a novel method that combines SAO (Subject-Action-Object) semantic structure with BERTopic to address this gap. The proposed method utilizes a deep pre-trained language model to establish a correlation matrix between scientific and technical literature and SAO-annotated technical terms. By leveraging the output of the language model, the method captures the underlying semantic relationships between the literature and technical terms, enabling a more comprehensive analysis. In a sequential manner, the method focuses on identifying the gap between science and technology. For scientific literature, a combination of OPTICS (Ordering Points to Identify the Clustering Structure) and DBSCAN (Density—Based Spatial Clustering of Application with Noise) algorithms is employed to extract technological topics. The advantage of this combined algorithm is its ability to perform outlier detection while clustering topics, without the need for complex parameter tuning. This feature contributes to its efficiency, making it a practical and effective approach for technology opportunity identification. In the context of extracting technological topics from patent literature, the study employs the GTM (Generative Topographic Mapping) Patent Map method to effectively identify and map technology vacuums. In the final stage, the method employs TF-IDF (Term Frequency–Inverse Document Frequency) cosine similarity to identify and filter scientifically feasible gaps between science and technology as potential technology opportunities. To validate the effectiveness of the proposed method, the study applies it to the domain of solar thermal power generation technology. The scientific and technical literature from 2013 to 2022 serves as an empirical case. The results demonstrate the applicability of the method, with scientific topics clustered into three classes and 22 outliers identified. In addition, the analysis of patent literature reveals four technology vacuums that effectively reflect areas with valuable information. By screening the gap between science and technology, the study identifies four potential technology opportunities, including coating technology based on metals, ceramics, and powders, vacuum powder insulation methods, tube and plate collector devices, molten salt materials based on chlorides and nitrates. Importantly, these opportunities align with the actual application of technologies in the field. In summary, this research enriches and extends the existing methods for technology opportunity identification by combining SAO semantic structure and BERTopic, as well as integrating the methods of OPTICS and DBSCAN. The combination of these approaches enhances the interpretability and efficiency of technology opportunity identification. The findings of the study provide valuable insights for professionals in relevant technology fields, enabling them to assess technology development trends and identify potential areas for innovation. In conclusion, the proposed method improves interpretability while maintaining efficiency. The empirical case in solar thermal power generation technology validates the effectiveness of the approach, leading to the identification of four potential technology opportunities. This research contributes to the field by providing a comprehensive and efficient method for technology opportunity identification, benefiting professionals in technology-related domains.

摘要： 借助科技差距概念识别技术机会已得到学术界关注，然而鲜有研究考虑科技文献深层语义特征的挖掘，进而提高技术机会识别结果的可解释性。研究提出SAO语义结构与BERTopic相结合的方法，基于深度预训练语言模型输出科技文献与带有SAO语义标注的技术术语之间的关联矩阵。相继地，在识别科技差距阶段，对于科学文献中技术主题的抽取，采用OPTICS和DBSCAN相结合算法，该算法能够在主题聚类的同时完成对离群点的检测，并且无需复杂的参数调试过程，具有一定的高效性；针对专利文献中技术主题的抽取，运用GTM专利地图的方法识别技术空白点；最后使用TF-IDF的余弦相似度识别并筛选出科学可行的科技差距作为潜在技术机会。研究选取太阳能热发电技术2013-2022年的科技文献为实证案例，检验方法的适用性与有效性，结果显示：论文主题聚类为3类，离群点22个，专利中能反映出有效信息的技术空白点4个，经过科技差距的筛选最终可得4个技术机会，分别是：基于金属、陶瓷和粉末的涂层技术、真空粉末绝热方法、管板式集热器设备和基于氯化物和硝酸盐的熔融盐材料，识别结果与技术实际应用情况相一致。研究丰富并扩展了技术机会识别方法，优化了已有技术机会识别方法的可解释性与高效性，研究结论可以为相关技术领域人员在判别技术发展态势时提供价值参考。