创新主体预测竞争对手或自身专利能否获批及获批速度,是一项重要的竞争情报工作.审批耗时会占用专利的有效期,有的发明专利审批耗时达8年甚至12年.审批耗时越长,对从申请日起算最长20年的专利有效期的折损就越大,对专利价值的影响也越大.本文旨在构建一种情报学方法,通过挖掘专利文献的系列特征形成预测模型,用于预测专利能否获批以及获批的速度.整个研究包括两个部分逻辑内容,即获批预测和速度预测.首先,应用相关分析与Cox比例风险回归模型对所选特征进行检验.其次,在此基础上,针对前面提取到的技术内容量属性、技术结构属性、技术功能属性、技术概念明确度属性以及申请人发明人属性等系列特征,使用辅助学习的方法,利用审查结果与审查周期的关联信息构建专利获批速度预测模型(patent approval time prediction model,MACP).研究结果显示,基于辅助任务组合的MACP较已有的基线模型表现更佳.由于MACP模型有效地学习与利用了更多的专利审查过程知识,降低了对数据量的依赖,能取得更好的预测效果.
An Approval Time Prediction Method Based on Patent Characteristics
Predicting the approval speed of their competitors'or their own patents is a crucial part of competitive intelli-gence analysis for innovation subjects.The examination time included the validity period of a patent.Some examiners took eight or even 12 years to approve a patent.This greatly reduces the validity period of a patent for up to 20 years from the date of application and has a considerable impact on its value.This study aimed to construct an information science meth-od by mining a series of features of patent literature to form a prediction model,which can be used to predict whether a pat-ent will be granted and the speed of approval.The research consisted of two logical parts:approval and speed prediction.First,correlation analysis and the Cox proportional risk regression model were applied to examine the selected characteris-tics.A patent approval time prediction model(MACP)based on the relation of examination decision and pendency was then constructed using the method of auxiliary learning,including the characteristics of technology internal capacity,tech-nology structure,technology function,technology concept clarity,applicant,and inventor.Experimental results show that the MACP based on the combination of auxiliary tasks outperformed the existing baseline model.Because the MACP mod-el can effectively learn and use more knowledge of the patent examination process,it can reduce the dependence on the amount of data and achieve a better prediction effect.
NETL
NSTL
万方数据
