Improving Mandarin Emotional Vocabulary in Application Writing Using Deep Learning Models

This paper first proposes a method for constructing an emotion dictionary based on the field of application writing. The evaluation method of user rating is utilized to determine the polarity of the target words, and the intensity of the emotional polarity of the target words is calculated at the same time, so as to construct an emotion dictionary based on the field of application essay writing. Then the CBOW model is improved, and some network structures are added under the original CBOW model to get the sentiment features, the semantic dependency path features of words are extracted through the Huffman semantic binary tree structure, and finally the bidirectional LSTM network is combined to propose the construction of a dynamic sentiment lexicon of Putonghua with the total features of the sentiment features, the semantic features, the information of the center word, and the distance of the words to the center word as the total features. The study applies emotion mining to application writing texts from learners from five types of institutions, and it is found that the writing texts from junior and senior high schools and higher institutions show significant differences in emotion tendencies. The study combines the indicators of affective tone, affective process and drive to realize the quantitative interpretation and comparison of learners’ affective tendencies in text writing from different institutions, and also fully confirms the good feasibility of the affective lexicon method proposed in this paper in the affective mining of textual materials for application writing.