Patent Analysis And Research On Garbage Sorting Based On BERT Model

With the further development of intellectual property rights as a strategic resources and core competitiveness of countries.With the increasing number of patent documents,how to quickly read and analyze a large number of patent documents has become the research direction of many experts and scholars at present.Nearly,with the deep learning,natural language processing technology(NLP)has made a new technological breakthrough,and language training model has become more and more perfect in terms of processing unstructured data,especially text processing and analysis.For the field of patent text,it is worth exploring to apply natural language processing technology to the processing and analysis of patent text data.This paper studies how to use natural language processing(NLP)to process and analyze patent text.First of all,this paper expounds in detail the patent analysis method,natural language processing technology,Bidirectional Encoder Representations from Transformers and related technologies and methods of patent auxiliary document generation.Secondly,referring to the idea and technology of text mining and automatic text abstract,this paper proposes a method based on the combination of BERT and LSTM neural network of large-scale language training model to realize the automatic generation of auxiliary documents of patent literature.The Encoder BERT layer using large-scale language training model to forecast the patent text feature extraction,training to solve the traditional processing model of the Transformer is difficult to solve the problem such as polysemy,could extract the deeper the linguistic features of information,get more granular text context said.Decoder layer uses twolayer LSTM neural network as text sequence output,extracts information from the information understood by the model to generate language,and realizes the automatic generation of documents.Add the Attention mechanism between encoder and decoder.Finally,this paper takes the patent in the field of garbage sorting technology as the research object of this experiment,and takes the invention content of the garbage sorting patent as the data set for experimental research.The experimental results show that the patent auxiliary documents can be used to summarize and reflect the generality,characteristics and technical composition of the patent.which can provide relevant technical information for patent readers and assist applicants to apply for patents in related technical fields.