An Important Scientific Research Mining Based On Scientific Papers

In recent years,the world has undergone earth-shaking changes.Scientific and technological personnel and innovation in scientific research achievements have become the most important force for social progress,common people's prosperity and national prosperity.For evaluation system of scientific papers in the past,mainly is the intelligence,researchers and experts and scholars as the leading to the summary of the important scientific and technological achievements,the evaluation result is not accurate and objective,expert artificial screening of all resources and research workload is very big,the cost is very high,it is difficult to have a fast and efficient evaluation results.By analyzing importance evaluation criteria for scientific papers in recent years,found that most of the analysis method is the use of scientific papers reference information of building mathematical model fitting the importance of scientific papers,and the use of information is mainly paper references,many methods are transplanted the social sciences,economic management maturity model and mathematical disciplines,and the use of machine learning methods to assess the importance of scientific papers articles is less.Therefore,this paper focuses on how to use multi-source information of scientific papers and statistical machine learning and other methods to build an objective,efficient and real-time mining model of important scientific papers,describes the directional collection algorithm of scientific papers,builds a graph structure model of scientific papers,and analyzes the research direction of hot spots.In scientific papers based on the statistical analysis of this thesis is about the importance of scientific papers dig,the paper references,downloads,views,factors such as social networks and news media combined into multi-source feature,after the publication of the early,constructed in the month of the temporal characteristics of multi-source and papers published after two years of multi-source feature,more exquisite and vividly depict the development and status of scientific paper early.To solve the problem of overfitting of time-series multi-source features combined with neural network,a new EW-Dropout algorithm was proposed to embed in the LSTM model to form the LSTM-EW optimization model.The LSTM-EW timing sequence optimization model and XGBoost state model constructed in this paper were fused to obtain 82.23% accuracy and 80.61% recall rate,and the recall rate effect was much higher than that of a single SVM,RF,XGBoost and LSTM-EW model,indicating that the fusion model based on timing features and state features can mine more important scientific paper achievements.This paper also provides a new idea for the evaluation of important paper results,reduces the workload of expert evaluation of scientific results,effectively uses the internet social information of scientific papers,and the results are more objective,efficient and accurate.