Topic Prediction Method Based on DTM-BiLSTM Model
ZHAO Lifei
GAO Quanli
SHAO Lianhe
JIN Lei
FENG Chen
YAN Hui
LUO Tongtong
Abstract：Topic models are widely used in intelligent recommendation,clustering analysis,topic evolution analysis and other fields to process different types of text data.However,traditional topic models do not consider the characteristics of text data time se-ries enough,resulting in the lack of time direction connection between topics in topic prediction.For this reason,this paper propos-es a topic prediction model based on the dynamic topic model and two-way short-term memory neural network.The dynamic topic model is used to classify and reduce the dimension of the chronological literature summary data,fully mining the evolution law of the data in time,and input the probability weight of the topic time to the BiLSTM model to predict the topic evolution in the future.Com-pared with the three control models(DTM-LSTM,DTM-GRU and DTM-BiLSTM),the dynamic theme model integrates the predic-tion model of short-term memory neural network,and has improved in the prediction evaluation indicators.The prediction result fit is 0.932 855,and the prediction accuracy of this algorithm is higher than that of DTM-GRU and DTM-LSTM models.
Keywords：dynamic topic modelsbi-directional long short-term memorytopic predictionnatural language processingpublic cultural servic
Publication Date：2025-09-20
Online Publishing Date：2025-12-23(First online date of this platform, not the publication date of the document)
Pages：8( 2489-2496 )