高校可转化专利识别模型构建——以人工智能领域为例

The Construction of the University Transferability Patent Recognition Model:A Case Analysis of Artificial Intelligence

[目的/意义]探索构建高校可转化专利识别模型,精准锁定存在转化价值的专利,对提升高校专利转化效率具有促进作用。[方法/过程]以人工智能领域为例,首先结合已有研究中使用的专利评估指标,从数据易获取性角度出发,确定了16个指标,并应用主成分分析探索指标之间的相关性,实现对重要指标的筛选;进而,将LDA模型与K-means算法结合,确定专利的技术主题,并与专利评估指标进行融合构建专利特征矩阵;最后,利用AdaBoost算法进行识别模型构建,并应用到对高校可转化专利的识别中,获取每件专利的可转化概率,仿照标准十分评估方法对专利进行划分,确定具备转化价值的专利。[结果/结论]结果显示,将专利技术主题与评估指标融合后,AdaBoost算法分类准确度提高了10%;通过对高校专利的识别,专利可转化概率呈对数常态分布,具备转化价值的专利比重为22.47%,验证了模型的有效性,为高校专利价值评估研究提供了新的研究思路,也为高校科技成果管理部门的专利运营及企业对高校核心专利的识别提供了实践方案。

[Purpose/significance]To explore the transferability patent recognition model of universities,and to accurately lock the patents with transfer value can promote the efficiency of patent transfer in universities.[Method/process]This paper takes artificial intelligence as an example.Firstly,based on the patent evaluation indicators in existing researches,a patent evaluation system consisting of 16 indicators was constructed from the perspective of accessibility of indicators.Furthermore,the principal component analysis was applied to explore the correlation between indicators and achieve the screening of important indicators.What’s more,the LDA model is combined with the K-means algorithm to obtain the technology topics of the patent and integrate with the patent evaluation indicators.Finally,the AdaBoost algorithm was used to build the recognition model,which was applied to the recognition of universities’transferability patents.The transferability probability of each patent was obtained,and the patents with transfer value were classified according to the standard score evaluation method.[Result/conclusion]After combining the patent technology topics with the evaluation indicators,the AdaBoost algorithm was applied for model training,and the accuracy of the classification algorithm was improved by 10%.Through the identification of transferability patents in universities,the proportion of the patents with transfer value is 22.47%,and the probability of transferability patents presents a logarithmic normal distribution.This model provides a new research idea for the evaluation of university patent value,as well as a practical scheme for the patent operation of university scientific and technological achievements management department and the identification of university core patents by enterprises.