侵权诉讼背景下标准必要专利价值分类识别体系构建

Construction of Standard Essential Patent Value Classification Recognition System Under the Background of Infringement Litigation

[目的/意义]基于机器学习算法,对行业标准专利构建多模态特征融合的自动分类筛选模型,探究侵权诉讼背景下标准必要专利价值分类指标体系。[方法/过程]首先利用美国专利商标局的发生侵权诉讼后的标准必要专利作为标记数据,将文本数据和指标数据进行降维融合后,建立基于机器学习中监督学习和半监督学习模型专利分类筛选模型,最后对数字创意产业的标准专利进行分类筛选。[结果/结论]基于机器学习中监督学习和半监督学习模型算法构建一套较为完整的多特征融合专利价值自动分类筛选模型。构建的4种模型在测试集上的平均F1值均在0.8以上,其中伪标签随机森林模型表现最优,平均F1值达到0.87106。

[Purpose/Significance]Based on machine learning algorithm,an automatic classification and screening model based on multi-modal feature fusion is constructed for industry standard patents.The research also explores a classification indicator system for the value of standard-essential patents in the context of infringement litigation.[Method/Process]First,standard necessary patents after infringement litigation in USPTO are used as marker data.Then,the text data and indicator data are integrated with dimensionality reduction,and the patent classification and screening model based on supervised and semi-supervised learning machine model is established.Finally,the standard patents of digital creative industry are classified and screened.[Result/Conclusion]The average F1 value of the four models constructed in this paper is above 0.8 on the test set,among which the pseudo-labeled random forest model has the best performance and the average F1 value reaches 0.87106.