Research On The Knowledge Mining Of Scientific And Technological Texts For "Question-Method" Utilization

The analysis and use of scientific and technological text content has always been a key issue in the study of information science.At present,science and technology have entered a stage of rapid development,and a large amount of scientific and technological literature has been accumulated in various fields.The traditional text mining method based on documents is not conducive to users' acquisition and utilization of knowledge content in scientific and technological literature.In order to make deeper and more effective use of scientific and technological texts,it is necessary to break through this mining scheme with a single text as the granularity,and from a more fine-grained perspective to the potential knowledge content of scientific and technological texts.The mining and utilization of scientific and technological texts play an important role in the development of science as a whole,especially the fine-grained and deep-level content such as research questions and solutions in scientific and technological texts that people focus on.These are important human knowledge accumulations.In view of this,this article focuses on the use of "question-method" related knowledge in scientific and technological papers to conduct research on knowledge mining of scientific and technological texts.It is specifically divided into two parts,including knowledge units and knowledge associations.For fine-grained knowledge mining in a specific field,they often correspond to entities and entity relationships.This article takes the field of artificial intelligence as an empirical object,and uses the abstract of scientific and technological literature as a data source to conduct research on the extraction of knowledge units and knowledge associations corresponding to "research questions" and "solutions".Comprehensively use a method based on a general neural network(such as Bi-LSTM)and a method based on a pre-trained language model(such as BERT,Sci Bert)to carry out on research questions and solutions entity identification and extraction research.And comprehensively compare the effects of each model on the entity recognition in the field of artificial intelligence.Then,taking the evolution of knowledge in the field of artificial intelligence as the application scenario,the best model is applied to the full set of abstracts in the field of artificial intelligence,and combined with the questions and method entities obtained from the statistical analysis of time factors,the distribution of questions and methods from different years,and the annual distribution of different questions and methods are analyzed from two perspectives.And the persistent and representative research questions and solutions in the field of artificial intelligence and the evolutionary trends of different questions and methods are found.Based on the results of entity recognition of questions and methods,the research on the knowledge correlation of scientific and technological texts in the field of artificial intelligence is carried out.Based on the supervised word embedding relation analogy and the unsupervised similarity calculation scheme,the question-question's high correlation and hierarchical relationship,the question-method's solving relationship,and the method-method's high similarity and hierarchical relationship are mined.Finally,based on the research results of entity relationship discovery,using knowledge question and answer as the application scenario in artificial intelligence domain,a specific research question instance is selected,and the visual application prototype design of the entity relationship discovery in the field of artificial intelligence is displayed.Then the relationship between questions and methods entities,the distribution and evolution of questions and methods are visually displayed.