| 王琲,杨坤,殷涛.基于BERTopic-LSTM模型的新兴技术主题预测与价值评估——以中国人形机器人为例[J].数字图书馆论坛,2025,21(12):48~58 |
| 基于BERTopic-LSTM模型的新兴技术主题预测与价值评估——以中国人形机器人为例 |
| Topic Prediction and Value Evaluation of Emerging Technologies Based on BERTopic-LSTM Model: Taking Chinese Humanoid Robot as an Example |
| 投稿时间：2025-10-16 |
| DOI：10.3772/j.issn.1673-2286.2025.12.006 |
| 中文关键词: 新兴技术；技术预测；BERTopic；LSTM；人形机器人 |
| 英文关键词: Emerging Technology; Technology Prediction; BERTopic; LSTM; Humanoid Robot |
| 基金项目:本研究得到教育部人文社会科学研究规划基金“‘知识重组-场景重构’情境下数字创新空间中集群行为的演化、涌现及调控研究”（编号：22YJA630104）资助。 |
| 作者 | 单位 | | 王琲 | 上海工程技术大学管理学院 | | 杨坤 | 上海工程技术大学管理学院 | | 殷涛 | 上海工程技术大学管理学院 |
|
| 摘要点击次数: 3 |
| 全文下载次数: 6 |
| 中文摘要: |
| 科学研判新兴技术的演化路径对国家战略科技布局具有重要的前瞻启示价值。针对现有研究细粒度主题捕捉能力不足、时序特征融合欠缺以及评估维度单一等局限，本研究构建了“技术主题识别—新兴技术预测—潜力价值评估”分析框架，旨在为技术优先排序与资源配置提供更为精准、可解释的决策依据。首先，提出考虑语义演化与时序融合的BERTopic-LSTM技术预测模型，克服传统模型难以兼顾文本深层语义与演化连续性的难题；其次，设计二维评估矩阵，透视新兴技术的成长动能与生态势能，系统识别其潜在战略价值；最后，以人形机器人领域为例进行实证分析。结果显示：BERTopic-LSTM模型在海量专利数据中识别出的技术主题更具细粒度，与其他模型对比，在均方根误差、平均绝对误差及R2等指标上均表现更优，预测结果与现实技术迭代轨迹高度契合。研究为政府与企业在复杂环境下的创新管理决策提供了科学的理论支撑与工具参考。 |
| 英文摘要: |
| Scientifically judging the evolution path of emerging technologies has important forward-looking enlightenment value for the national strategic layout of science and technology. In view of the limitations of existing research in fine-grained topic capture ability, lack of temporal feature fusion, and single evaluation dimension, this study constructs an analysis framework of “technology topic identification?emerging technology prediction-potential value evaluation”, aiming to provide more accurate and interpretable decision-making basis for technology priority ranking and resource allocation. Firstly,a BERTopic-LSTM technology prediction model considering semantic evolution and temporal fusion is proposed to overcome the problem that traditional models are difficult to take into account the deep semantics and evolution continuity of text. Secondly, a two-dimensional evaluation matrix is designed to see through the growth impetus and ecological potential inclination of emerging technologies, and systematically identify their potential strategic value. Finally, an empirical analysis is carried out in the field of humanoid robot. The results show that the technical topics identified by the BERTopic-LSTM model in massive patent data are more fine-grained. Compared with other models, it performs better in RMSE, MAE, and R2, and the prediction results are highly consistent with the actual technology iteration trajectory. The research provides scientific theoretical support and tool reference for the innovation management decision-making of government and enterprises in complex environment. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
