利用词嵌入模型实现基于网站访问日志的专利聚类研究

摘要

[目的/意义]专利信息是人类科学技术进步的结晶,随着社会的发展,专利信息将为促进科技创新发挥日益重要的作用.利用聚类技术可以将海量专利信息进行自动分类,在实现信息有序归并管理的同时,有助于用户高效而全面的获取相关技术领域中的集成专利信息,具有重要的现实意义,传统聚类研究方法效率与准确度存在不足.[方法/过程]本文通过对专利信息服务网站(中国科学院知识产权网)访问日志数据的清洗与分析,生成专利信息点击序列,基于深度学习词嵌入模型,设计了PatentFreq2Vec模型,计算得出专利关联信息.[结果/结论]利用 PatentFreq2Vec模型分析计算访问日志数据,能够得到关联专利信息,实现专利聚类,且聚类准确度高于传统方法.%[Purpose/Significance]Patent information is the fruit of the progress of science and technology.With the development of society, patent information will play an increasingly important role in promoting scientific and technolog-ical innovation.Through patent clustering analysis, it is possible to aggregate isolated information according to different ag-gregation degree, so that they can be transformed from ordinary information to valuable Patent Competitive intelligence.The traditional clustering methods have some efficiency and accuracy problems.[Method/Process]Based on cleaning and anal-ysis access log data of the patent information service website(Intellectual property network of the Chinese Academy of Sci-ences), the sequence data of patent clicking were generated and input into the PatentFreq2Vec model based on word em-bedding to obtaine patent related information with the learning algorithm.[Result/Conclusion]This could cluster the pa-tents and improve accuracy of the patent clustering.