| 齐世杰,串丽敏,张辉,姚茹,贾倩,赵静娟.基于特征测度和SciBERT模型的突破性科学创新主题识别研究——以农业机器人领域为例[J].数字图书馆论坛,2025,21(9):16~27 |
| 基于特征测度和SciBERT模型的突破性科学创新主题识别研究——以农业机器人领域为例 |
| Breakthrough Scientific Innovation Topic Identification Based on Feature Measure and SciBERT Model: A Case Study of Agricultural Robots Field |
| 投稿时间：2025-07-27 |
| DOI：10.3772/j.issn.1673-2286.2025.09.003 |
| 中文关键词: 突破性科学创新；主题识别；农业机器人；机器学习；跨学科研究；SciBERT；K-means |
| 英文关键词: Breakthrough Scientific Innovation; Topic Identification; Agricultural Robot; Machine Learning; Interdisciplinary Research; SciBERT; K-means |
| 基金项目:本研究得到北京市农林科学院科技创新能力建设专项“智库型农业情报研究与服务能力提升”（编号：KJCX20230208）、北京市农林科学院科技创新能力建设专项“面向科研管理的情报研究与服务能力提升”（编号：KJCX20230210）资助。 |
| 作者 | 单位 | | 齐世杰 | 北京市农林科学院数据科学与农业经济研究所 | | 串丽敏 | 北京市农林科学院数据科学与农业经济研究所 | | 张辉 | 北京市农林科学院数据科学与农业经济研究所 | | 姚茹 | 北京市农林科学院数据科学与农业经济研究所 | | 贾倩 | 北京市农林科学院数据科学与农业经济研究所 | | 赵静娟 | 北京市农林科学院数据科学与农业经济研究所 |
|
| 摘要点击次数: 52 |
| 全文下载次数: 57 |
| 中文摘要: |
| 突破性科学创新是推动科技进步和社会经济发展的核心动力，准确识别突破性科学创新，对于开辟科技创新的新赛道、实现跨越式发展具有重要意义。首先从突破性科学创新特征入手，综合考虑论文的多样性、均衡性和差异性，利用True Diversity测度指标和学科共现网络进行潜力论文的测度与筛选；然后采用SciBERT+K-means+UMAP方法进行主题建模与可视化，从新颖性、内容变革性、影响力、突变性4个维度构建多特征的突破性科学创新指标体系，融合多种文本挖掘方法识别突破性科学创新主题；最后以农业机器人领域为例，识别出4个突破性科学创新主题：多模态感知与自适应智能决策、机器人仿生驱动与异质能源耦合、群体协同与云边端架构技术、高精度轨迹控制与高动态响应，验证了模型的有效性和可靠性。研究不仅为突破性科学创新主题识别提供了新思路，还为智慧农业领域的原始创新提供了可参考的依据。 |
| 英文摘要: |
| Breakthrough scientific innovation is the core driving force behind technological progress and socio-economic development. Accurately identifying breakthrough scientific innovations is of great significance for opening up new tracks of scientific and technological innovation and achieving leapfrog development. Starting from the characteristics of breakthrough scientific innovations in scientific paper data, we first comprehensively consider the diversity, balance, and difference of papers, using the True Diversity measurement index and disciplinary co-occurrence network to measure and screen potential papers. Then, we employ the SciBERT+K-means+UMAP method for topic modeling and visualization. From the four dimensions of novelty, content transformation, influence, and abruptness, we construct a multi-feature breakthrough scientific innovation indicator system, integrating various text mining methods to identify breakthrough scientific innovation topics. Taking the field of agricultural robots as an example, our breakthrough scientific innovation topics are ultimately identified: multimodal perception and adaptive intelligent decision-making, bio-inspired actuation and heterogeneous energy coupling for robots, swarm collaboration and cloud-edge-end architectural technology, and high-precision trajectory control and high-dynamic response. The effectiveness and reliability of the model are verified. The research not only provides new ideas for identifying breakthrough scientific innovation topics, but also provides a reference basis for original innovation in the field of smart agriculture. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
