首页|融合权利要求语义特征的专利价值早期预测研究

融合权利要求语义特征的专利价值早期预测研究

扫码查看
原文链接
  • 万方数据
  • 维普
[研究目的]探索有效的方法实现对专利价值的早期预测既可为后续高价值专利培育提供更多机会,也可对问题专利等低价值专利进行早期识别,减少公共资源虚耗,对推动知识产权高质量发展至关重要.[研究方法]聚焦现有研究鲜少关注的专利权力要求书,提出一种融合权利要求语义特征的专利价值早期预测方法.首先,对权利要求文本进行解析并提取语义特征;然后融合权利要求语义特征构建专利价值评估指标体系并运用皮尔逊相关性分析方法进行指标优化;最后,结合CatBoost分类模型和SHAP方法实现高价值专利的早期挖掘和低价值专利的早期识别.[研究结论]结果表明,构建的专利价值早期预测模型分类准确率达到90.79%,相较于传统机器学习方法具有准确性更高、解释性更好、适用性更广的优势;首次引入的权利要求语义特征对于预测专利价值贡献显著,融合该特征的指标体系具有一定的优越性.
On Early Prediction of Patent Value by Fusing Semantic Features of Claims
[Research purpose]Investigating efficient methods to achieve early patent value prediction can not only increase the opportuni-ties for the subsequent cultivation of high-value patents,but also can produce early identification of low-value patents,such as questiona-ble patents,to lessen the depletion of public resources,which is essential to support the high-quality development of intellectual property.[Research method]An early prediction method of patent value by fusing semantic features of claims is proposed.The semantic features are initially extracted from the parsed text of patent claims.Then,an index system for early patent value prediction is built with an integra-tion of the semantic features of claims and then optimized with pearson correlation analysis.Finally,the CatBoost classification model and the SHAP method are combined to realize the early mining of high-value patents and the early identification of low-value patents.[Re-search conclusion]The results show that the early prediction model of patent value built in this study has a classification accuracy of 90.79%,which has the advantages of higher accuracy,better interpretation,and wider applicability compared with the conventional ma-chine learning methods.The semantic features of claims have a significant impact on patent value prediction,and the index system built by integrating this feature has some advantages.

patent valuepatent claimsemantic featuresprediction modelevaluation index systemCatBoostSHAPinterpretable machine learning

付姣、梁丽芝

展开 >

湘潭大学公共管理学院 湘潭 411105

专利价值 权利要求 语义特征 预测模型 评估指标体系 CatBoost SHAP 可解释机器学习

湖南省教育厅科研项目湖南省研究生科研创新项目湘潭大学研究生科研创新项目

21B0161CX20220580XDCX2022Y016

2024

10.3969/j.issn.1002-1965.2024.02.025

情报杂志
陕西省科学技术信息研究所

情报杂志

CSTPCDCSSCICHSSCD北大核心
影响因子：1.502
ISSN：1002-1965
年,卷(期)：2024.43(2)
  • 34