Research On Metaphor Recognition And Interpretation Based On Deep Learning

Metaphors can be seen everywhere in daily life.Studies on cognitive linguistic show that metaphor is also an important cognitive method.With the topic of artificial intelligence continues to heat up,natural language processing research has become a hot spot of common concern in academia and industry.Metaphor computation is gradually attracting the attention of scholars in the field of natural language processing.It is also an important issue that cannot be avoided in natural language processing.In recent years,the deep learning method has developed rapidly,and the deep neural network has obvious advantages in feature learning.In view of this,this paper mainly studies the metaphor recognition and interpretation based on the deep learning method.Firstly,we propose a method of metaphor recognition based on convolutional neural network and SVM classifier.This method is mainly aimed at Chinese nominal metaphors.We extracted feature by convolutional neural networks and SVM classifiers.In order to solve the problem of polysemy,the word vector and part-of-speech feature are proposed as the input of neural network.For the incompleteness of features in the pooling layer of convolutional neural network,the combining of Max-pooling and Mean-pooling is proposed.It improves the accuracy of feature extraction and the accuracy of metaphor recognition.The results show that the accuracy of the metaphor recognition by convolutional neural networks and SVM classifiers is higher than that of directly using convolutional neural networks.Moreover,the experiment of metaphor recognition on English datasets further proves the effectiveness of this method in metaphor recognition.Furthermore,in view of the direct use of convolutional neural networks to extract features,long-distance dependence and semantic features cannot be captured,which affects the accuracy of feature extraction.To solve this problem,this paper proposes a method based on Transformer framework,which improves the parallelism of the system and the training time.The feature through the attention calculation of the Transformer framework,and applies it to the fully connected layer for classification and recognition.The vector matrix of the sentence and the position feature as the input sequence of the network structure,the long distance feature in thesentence can be obtained,and the forward network uses the convolutional neural network in the Transformer framework to make the extracted features more accurate.Experiments show that this method has better metaphor recognition effect and is superior to the recognition algorithm based on convolutional neural network and SVM classifier.Finally,The metaphor understanding method based on Seq2 seq framework is constructed.Firstly,We construct a metaphor understanding corpus,which mainly consists of metaphorical sentences and metaphorical understanding sentences.Then,the interpretation of metaphor through the framework of the encoder-decoder.In the model,the bidirectional LSTM network model and attention mechanism are selected,it improve the effect of the metaphor understanding model.