Technology Topic Prediction Research Based On LSTM

Since entering the 21 st century,the world has entered a period of rapid development of scientific and technological innovation.The advancement of science and technology has greatly promoted the development and progress of society.It is of great significance to track the progress of scientific research and predict the future research trends of science and technology.The traditional technology prediction method is mainly based on Delphi's expert discussion form,its existence is dependent on technical experts,and the cost of manual processing is high.The machine learning-based prediction method brings hope to solve many shortcomings of traditional subjective methods in technology prediction.However,in the face of massive data,using effective dimensionality reduction methods to characterize scientific and technological literature,the temporal evolution of modeling technology topics has become the core issue in technology topic prediction.This paper proposes a LSTM-based technology topic prediction method,which improves the existing methods from the following two aspects: Firstly,the Gibbs sampling process in the LDA model is improved on the acquisition side of the technology topic to improve the topic mining model.The accuracy and stability of the document data to achieve more efficient dimensionality reduction.This step is based on the LDA model to obtain the initial set of subject words.According to Word2 vec,the words with similar semantics are extracted from the initial subject word set,and the initial topic lexical similarity and vector adjacency are redistributed according to different weights to improve the Gibbs sampling.Secondly,in the technical prediction,it proposes to add three kinds of features: technology topic strength,keyword frequency,and document quantity to improve the model prediction accuracy.Considering the good processing ability of LSTM for sequence data and the memory function for long-term information,this paper constructs a technology topic prediction model based on LSTM.By comparing the prediction results of this model with those of other machine learning models and LSTM model with single feature,it is found that the accuracy of the modeling method adopted in this paper has been improved in different degrees compared with that of other models.