Research On The Sentiment Classification Method Of Poetry Based On The Features Of Imagery

Sentiment analysis is an important subject in the field of natural language processing,but in the current research related to sentiment analysis,poetry text has rarely been chosen as research object.Chinese poetry,as the ideological crystallization of the Chinese nation,its related emotional research work is conducive to strengthening cultural education and promoting traditional Chinese culture.The language of the poetry is concise,and its emotion is usually carried in the imagery words.Imagery words combine objective images and subjective emotions,and contain metaphorical thinking and cultural characteristics.Currently,none of the methods for poetry sentiment analysis deeply integrate the characteristics of imagery words.we proposed a scheme for fusing the features of imagery words with mainstream deep neural network architecture for poetry sentiment analysis.The main innovative work is as follows:1.At present,there is no public attribute database of poetry imagery words,so this paper constructed the imagery word lexicon and the cultural cognitive attribute database of imagery words in a semi-manual way,and conducted follow-up research based on this.2.The construction of the image language of poetry is based on people’s perception and association of the natural attributes and characteristics of objective things under the background of specific cultural cognition.Therefore,this research started with the attributes of imagery words in poetry,and used the knowledge of the cultural cognitive attributes of imagery words as the source of cognitive information contained in the imagery words,and constructed a poetic text sentiment classification model based on the knowledge of the cognitive attributes of imagery words.The model used the attention mechanism to extract the contextual features of poetry related to the cognitive attributes of image words and takes them as key information to judge the sentiment polarity of the input text.The experiments proved that the integration of image cultural cognitive attribute information could improve the performance of sentiment classification model.3.Combining image depiction with emotional depiction,imagery words in poetry are objective things that incorporate subjective emotions.Language is abstract,while the image expression of objective things is more specific than language expression.Therefore,this research further introduced the relevant visual image modal information of imagery words in poetry,proposed a poetic multimodal sentiment analysis model based on imagery words.The model used a multi-headed self-attention mechanism to explore the correlation information between poetry text,imagery words,and image features related to imagery words,and explored the characteristics of poetry imagery in multiple dimensions.The experimental results showed that our model was effective and universal for the sentiment classification task of poetry.In summary,this research constructed a new imagery word database,and combined with the emotional characteristics of poetry text,constructed the text modal poetry sentiment classification model and multi-modal poetry sentiment classification model based on the imagery word characteristics,which also provided a new idea for sentiment analysis of other literary genres.