摘要
为研究专利匹配算法以提高专利匹配准确率并解决专利检索、专利推荐、专利地图等专利服务关联性不足的问题,文中提出一种基于双向表征编码器和注意力机制(BERT+ATT)以及基于密度聚类(DBSCAN)的专利匹配算法,首先使用BERT+ATT将专利文本转换成专利向量,然后采用DBSCAN对专利向量进行聚类,最后根据聚类结果和专利服务算法导出相应的专利检索、专利推荐、专利地图结果。基于BERT+ATT专利匹配算法的综合评价分数百分比为87.64%,相比于Word2Vec的75.93%提升了11.71%,相比于BERT的83.30%提升了4.34%。文中提出方法有助于分析和研究长三角地区的专利情况。
The patent matching algorithm is introduced to solve the problem of insufficient correlation among patent services such as patent search,patent recommendation and patent map.A patent matching algorithm based on BERT with attention mechanism(BERT+ATT)and DBSCAN is proposed.Firstly,a patent text is converted to a patent vector through BERT+ATT.Then the cluster which the patent vector belongs to is found by DBSCAN.Finally,the corresponding patent search,patent recommendation,and patent recommendation,and patent map results are derived to ensure the relationship between patent services.The comprehensive evaluation score of the patent matching algorithm based on BERT+ATT is 87.64%,which is 11.71%higher than 75.93% of Word2Vec,and 4.34%compared with 83.30% of BERT.This method is helpful for analyzing and studying patents in the Yangtze River Delta region.
作者
曹旭友
周志平
王利
赵卫东
CAO Xu-you;ZHOU Zhi-ping;WANG Li;ZHAO Wei-dong(College of Electronics and Information Engineering,Tongji University,Shanghai 201804,China)
出处
《信息技术》
2020年第3期1-5,12,共6页
Information Technology
基金
国家重点研发计划(2017YFB1401600)
