A DCRC MODEL FOR TEXT CLASSIFICATION

摘要

TRADITIONAL TEXT CLASSIFICATION MODELS HAVE SOME DRAWBACKS, SUCH AS THE INABILITY OF THE MODEL TO FOCUS ON IMPORTANT PARTS OF THE TEXT CONTEXTUAL INFORMATION IN TEXT PROCESSING. TO SOLVE THIS PROBLEM, WE FUSE THE LONG AND SHORT-TERM MEMORY NETWORK BIGRU WITH A CONVOLUTIONAL NEURAL NETWORK TO RECEIVE TEXT SEQUENCE INPUT TO REDUCE THE DIMENSIONALITY OF THE INPUT SEQUENCE AND TO REDUCE THE LOSS OF TEXT FEATURES BASED ON THE LENGTH AND CONTEXT DEPENDENCY OF THE INPUT TEXT SEQUENCE. CONSIDERING THE EXTRACTION OF IMPORTANT FEATURES OF THE TEXT, WE CHOOSE THE LONG AND SHORT-TERM MEMORY NETWORK BILSTM TO CAPTURE THE MAIN FEATURES OF THE TEXT AND THUS REDUCE THE LOSS OF FEATURES. FINALLY, WE PROPOSE A BIGRU-CNN-BILSTM MODEL (DCRC MODEL) BASED ON CNN, GRU AND LSTM, WHICH IS TRAINED AND VALIDATED ON THE THUCNEWS AND TOUTIAO NEWS DATASETS. THE MODEL OUTPERFORMED THE TRADITIONAL MODEL IN TERMS OF ACCURACY, RECALL AND F1 SCORE AFTER EXPERIMENTAL COMPARISON.