Comparative Analysis of Hot Technical Methods in Papers and Patents from the Perspective of Knowledge Evolution:A Case Study in the Field of Artificial Intelligence and Natural Language Processing
The research outcomes in the field of artificial intelligence transition and circulate between academic research and industrial practice.To uncover the distribution and evolutionary trends of technical methods in natural language processing(NLP)within papers and patents,a comparative analysis is conducted based on the perspective of knowledge evolution.This paper analyzes 39 025 NLP papers from the arXiv dataset spanning January 2013 to December 2022,and 14 654 patents'information retrieved from the incoPat patent database by using the search method of"keywords+IPC classification number".Using the Wikipedia category tree,an ontology of methods is constructed to map the papers and patents to the domain ontology.A paired sample t-test is employed to verify the lag in knowledge evolution between the two sources.The findings indicate that the sampled papers and patents share common technical methods,including convolutional neural networks(CNN),recurrent neural networks(RNN),long short-term memory(LSTM),attention mechanisms,and pretrained language models.Among these,the first three represent relatively mature neural network architectures,while the attention mechanism has rapidly developed alongside these neural network architectures.Patent technical methods tend to lag behind papers by approximately 1 to 2 years.
artificial intelligencenatural language processing(NLP)knowledge evolutionneural networkstechnical methodsbibliometricspatents analysis
NETL
NSTL
万方数据
