基于专利数据应用LDA和N-BEATS组合方法的技术主题预测研究  

Technology Themes Prediction Based on Combination of LDA and N-BEATS Methods Applied to Patent Data

作　　者：吴雷[1] 杜文研 林超然[1] WU Lei;DU WenYan;LIN ChaoRan(School of Economics and Management,Harbin Engineering University,Harbin 150001,P.R.China)

机构地区：[1]哈尔滨工程大学经济管理学院,哈尔滨150001

出　　处：《数字图书馆论坛》2023年第11期62-73,共12页Digital Library Forum

基　　金：国家社会科学基金一般项目“‘双链融合’视角下IC产业关键核心技术甄别及突破研究”(编号:23BGL076)资助。

摘　　要：预测技术主题未来热点,有助于企业在技术层面判别现状、识别未来技术方向并提前规划战略布局。提出LDA和N-BEATS组合方法,运用LDA模型提取专利文献数据的技术主题,引入N-BEATS网络模型分析各技术主题专利数量的时间序列,发挥其分析可解释性时间序列的优势,在预测模型中加入技术研发活动周期性模块,并以芯片技术为例,运用该组合方法预测产业的技术主题和未来趋势。对比实验中LDA和N-BEATS组合方法的预测精度高于LDA-LSTM、IPC-N-BEATS和IPC-LSTM三种基准方法。案例结果表明,未来芯片产业研发热点是电子级树脂、蚀刻机、芯片封装、芯片键合、抛光液。Predicting the future hot topics of technology themes can enable enterprises to distinguish the current situation,identify future technological directions,and plan their strategic layout in advance at the technical level.This study proposes a combination method of LDA and N-BEATS,which uses the LDA model to extract technical topics from patent literature data.The N-BEATS network model is introduced to analyze the time series of the number of patents for each technical topic,leveraging its advantage in analyzing interpretable time series.The periodic module of technological research and development activities is added to the prediction model,and the chip technology is used as an example to predict the technical themes and future trends of the industry using this combination method.The prediction accuracy of the LDA and N-BEATS combination method in the comparative experiment is higher than that of the three benchmark methods:LDA-LSTM,IPC-N-BEATS,and IPC-LSTM.The case results indicate that the future research and development hotspots in the chip industry are electronic grade resins,etching machines,chip packaging,chip bonding,and polishing fluids.

关 键 词：LDA N-BEATS网络模型 深度学习 芯片产业 技术预测 

分 类 号：G255[文化科学—图书馆学]