A New Method Of Metaphor Recognition For A-is-B Model In Chinese Sentences

The expression of metaphor frequently appears in the language used in daily life and in various situations.It is also a major challenge in natural language processing.If metaphor recognition technology cannot be advanced,the development of the field of natural language processing will also be limited,and because the meta-sentences of the A-is-B mode appear more flexible,the meta-recognition of the A-is-B mode is another difficulty in metaphor recognition.At present,the research of A-is-B model metaphor recognition has the following aspects: First,the scope of source domain words is more extensive.In the Ais-B model metaphor,the scope of both A and B is very broad.Second,context is more sensitive.An A-isB model metaphor sentence will have different meanings in different contexts.Third,the influence of pronouns.In A-is-B model metaphor sentences,A is often a pronoun,and it is necessary to convert the pronoun into a specific noun or noun phrase when identifying metaphors.Fourth,the existence of death metaphors.Some source domain words are too prominent because of their metaphorical meaning.People have assumed that the source domain words are inherently metaphorical in the process of using them.This paper selects the A-is-B pattern metaphor sentence in Chinese text as the research object.Aiming at the difficulty of A-is-B pattern metaphor recognition,a neural network classifier method based on LSTM and multi-feature fusion is proposed.Automatic recognition of metaphorical sentences in A-is-B mode.The main research contents of this article are as follows:(1)Feature extraction of A-is-B metaphorical sentences.Features were extracted from sentences using four methods based on the superordinate relation database,sentence model,class word,and Word2 Vec similarity.In the superordinate relation database,the superordinate relation in "Synonym Cilin" has been expanded,and the superordinate relation has been discriminated by using recursion;In sentence models,two matching methods of non-metaphor sentence models and metaphorical sentence models are proposed;In class words,a method of iteratively obtaining class words based on seeds was proposed,and the class words were expanded by mining the internal information of seeds;In Word2 vec,the method of extracting ending words is used to improve the rationality of similarity judgment.(2)Classification of A-is-B metaphorical sentences.A neural network classifier algorithm based on LSTM and multi-feature fusion is used.This method uses the LSTM neural network to extract features for the entire sentence.This feature is combined with the six features obtained by the four feature extraction methods in this paper,using a fully connected neural network.The network generates the final classification results,and then compares the labeled real results to adjust the parameters of the neural network part of the algorithm.The experimental results show that compared with the method using the SVM classifier and the LSTM classifier,the neural network method based on LSTM and multi-feature fusion has better accuracy and recall rates,which are 96.7% and 93.1%,respectively,but predict a Sentences are more time consuming.According to the analysis of the experimental results in this paper,the improved method has achieved good results.