CNC System Field Technical Term Recognition Based On Deep Transfer Learning

In recent years,with the proposal of the technological paradigm such as Industry 4.0 and Internet+,scientific and technological innovation has developed rapidly.As a strategic technology,CNC system technology undoubtedly belongs to the national core technology industry,and vigorously develops CNC machining technology is important for a country with strong manufacturing.Under this background,the study of extracting emerging technical terms,forecasting trends in the field,is of great significance for countries and enterprises to develop strategic development plans.As patent data constitute the latest source of technical intelligence,it has been widely used in emerging technologies foresight.Although patent documents are easy to use,the technological term is difficult to mine and extract.Lack of terminology labeling is the critical issue.Therefore,how to extract technical terms for patent data is the focus study of this thesis.In view of the shortcomings of the existing research,this thesis first introduces the idea of deep transfer learning,based on the named entity recognition technology,constructs the overall scheme of emerging term recognition,technology category division and patent trend analysis in the field of computer numerical control system;then,this thesis is based on language model and named entity recognition model for transfer learning in technology term identification,using mature public domain source data,applying Bi-LSTM(bidirectional long-term memory)model to achieve cross-domain transfer,effectively identifying technical terms and filtering high-frequency non-term word strings;By constructing the term word vector,WMD(word mover's distance)technology is used to calculate the document similarity and this thesis applies K-means to classify the technical categories of patent documents and technical terms.The clustering results are presented in term phrase form,which is more accurate,understandable and easy to interpret.Based on the above steps,this thesis collects patent data in the field of computer numerical control(CNC)systems from 2013 to 2018.By transferring the existing knowledge of the source data of the journalism domain to the target data of the CNC system domain,the problem of lack of labeling in patent data is solved,and the CNC system field is divided into five categories with the term word vector and document clustering: hardware,software,technology,network,and intelligence.This thesis combines the methods of patent analysis,integrates the classified technical categories,and comprehensively analyzes the development trend of the computer numerical control system field from 2013 to 2018.