A Metaphor Recognition Model based on LSTM and Keyword Similarity Computation

Metaphor, as a common figure of speech, can reflect literary grace and profoundness to a certain extent, and is an important indicator in article appreciation. However, at present, there is no perfect application model in automatic recognition. According to the three structures and basic principles of metaphorical sentences, this paper designs a metaphorical sentence recognition model based on LSTM and keyword similarity from the perspective of the sentence itself and its syntactic structure. It solves the problems that LSTM model cannot quickly extract the feature structure of metaphorical sentences and keyword similarity computation ignores the context semantics. The word vector and part of speech tagging are input into the multi-input cyclic neural network, and a recognition method based on LSTM model is designed. On the basis of the metaphorical comparison theory, by using the extracted metaphor tenor and vehicle, this paper improves the computation method of word similarity in CNKI, and designs a recognition method based on keyword similarity. The two, LSTM and keyword similarity, are integrated, with the comprehensive consideration of metaphorical sentences in structure, semantics and context. Finally, the model is verified in the collected corpus files, thus achieving high accuracy.