结合高阶神经元的深度神经网络用于专利可转让性评价——以电子信息技术领域为例

Patent Transferability Evaluation Based on Deep Neural Network with High-order Neuron

[目的/意义]从交易视角评价专利可转让性,侧面评估专利价值及筛选可交易高价值专利。[方法/过程]基于专利价值评估指标,从技术和法律两个维度选取专利可转让性评价内部指标,基于交易视角中专利权人特征设计专利可转让性评价外部指标,结合高阶神经元将深度神经网络方法应用于专利可转让性评价。[结果/结论]结果表明,专利可转让性评价模型相比传统的BP神经网络方法和仅使用高阶神经元的方法精度更高,F1值达到86.72%;因其可区分通过交易实现价值的潜在专利,在大规模专利可转让性评价实际应用中具有可行性和适用性。

[Purpose/Significance]From the perspective of transaction,we can evaluate the transferability of patents,evaluate the value of patents and screen high-value tradable patents.[Method/Process]Based on the evaluation index of patent value,the internal index of patent transferability evaluation was selected from two dimensions of technology and law.The external index of patent transferability evaluation was designed based on the characteristics of patentee in the perspective of transaction.The deep neural network method was applied to patent transferability evaluation combined with high-order neuron.[Result/Conclusion]Compared with the traditional BP neural network method and the method using only higher-order neurons,the accuracy of patent transferability evaluation model is higher,and the F1 value is 86.71%.Because it can distinguish the potential patents that realize value through transaction,it has feasibility and applicability in the practical application of large-scale patent transferability evaluation.