Scientific prediction on core patents is of importance to the strategic layout of an enterprise, which should be achieved on the basis of technosphere subdivision. The Louvain method is utilized to subdivide the patent network based on the similarity matrix, and then a predictive model for core patent discovery using a support vector machine (SVM) is constructed. This is composed of five indicators: citation counts in four years, patent families, patent width, patent claims, and science links. We carry out a demonstration of core patent prediction in the field of OLED. The model under the subdivision condition is compared with that under the non-subdivision condition, and the predictive effect of the model based on SVM is also compared with those using other algorithms, such as logistic, ra-dial basis function neural network, K-nearest neighbors, and Bayes discriminant. The result shows that the model based on technosphere subdivision is superior. Meanwhile, the impact of the indicator number on the prediction result is discussed, and it is also proven that the current model with five indicators is the most scientific and effective.%核心专利的科学预测对于企业技术战略性布局具有重要意义,而要实现这一目标需建立在技术领域细分的基础上.论文在构建专利相似性矩阵的基础上,利用Louvain社团发现算法对专利网络进行了领域细分.在细分视角下,利用"四年内被引频次、同族专利数、专利宽度、权利要求数、科学关联度"五个指标构建了基于支持向量机的核心专利预测模型.以OLED领域核心专利预测为例进行实证研究,从"是否进行技术领域细分"和"与其他常用分类预测方法区别"两个视角进行了比较分析,结果表明本文提出的基于技术领域细分视角的核心专利预测模型在预测效果上具有一定比较优势.与此同时,论文还讨论了指标个数遴选对于预测结果的影响,结果亦表明当前建立的五个指标预测模型相对而言最为科学合理.
展开▼
