The Theme Analysis Of Science Foundation Based On LDA Model

The data of science foundation project,similar to the literature data,is standardized and easy to acquire.In recent years there are much research on the science foundation project,including performance evaluation,regional distribution,input-output,research hotpoint and development trend etc.These methods are basically based on the methods of statistics and econometrics.In the research on topic and development trend,these methods can not find the hidden information and make the research results more macroscopic,with lacking of in-depth study and limitation.LDA model is a probabilistic generative model method of text mining,which can be used to find hidden information.It has been widely used in text clustering,text classification,evolution,policy analysis,recommendation etc.Applying the LDA model into the theme analysis of science foundation has important significance for the researchers and managers to understand research hotspot and development trend of science foundation project.This paper applies the LDA model into the data of the national natural science foundation project which are approved time interval between 2003-2012.Then it study science foundation theme and evolution process.Finally combined with the current situation,the result of evolution are analyzed in detail.The details are follows:(1)This paper summerizes the situation of science foundation and the LDA model through the literature,and then analyzes and compares the four theme analysis method.At the same time,the LDA model has feasibilities and advantages compared with the other three theme analysis method.Finally this paper put forward the complete process of the theme research and analysis of science foundation project based on LDA model.Meanwhile this paper refines and studies the key issues(including Gibbs parameter estimation method,optimal topic number determination method,post discrete theme evolution analysis method and theme measurement method and so on).(2)After selecting approved time interval of science fund data between 2003-2012,this paper use the LDA model to carry out the theme evolution analysis and then identify science foundation theme(supply chain management,energy planning,capital supervision,audit system of financial engineering,regional ecological network,brand management,mode of training team,the theory and method of prediction,investment,program and control of production)through the theme of filtering.It not only provide a reference for research specialist staff,but also provide decision support to formulate appropriate policies for scientific management department.(3)Through the LDA modeling to obtain 10 themes of the past ten years,this paper set the threshold to find that supply chain management and regional ecological network are identified as stable hot research themes,brand management and investment are stable research themes,the theory and method of prediction and mode of training team are new research hotspots.Then selecting the theme of supply chain management,this paper calculate KL divergence of the adjacent time window of the theme to analyze the content evolution.Finally through the curve of supply chain management and brand management which use the themes strength,the analysis of the heat of the two themes will continue to grow.The research result shows that applying the LDA model into the theme analysis of science foundation project make data analysis of science foundation project timely,accurate and efficient.It opens up a new path of theme analysis of science foundation.