Research On Implicit Sentiment Analysis Based On Tree Recurrent Neutral Network

With the rapid development of social media,text data generated by more and more social media platforms,such as microblog,Facebook,Instagram,has the characteristics of fast speed,wide area and large amount of data,which provides a lot of data support for today’s text analysis work,has gained significant popularity among the social network services.In sentiment analysis,explicit or implicit emotions are classified according to whether explicit emotion words are included.Due to the implicitness of implicit sentiment expression,the traditional sentiment analysis model can not be well applied in the field of implicit sentiment classification.There are various forms of implicit sentiment.According to grammar and semantics,implicit emotional atmosphere can be divided into four types: factual,rhetorical,satirical and metaphorical.For the implicit sentiment recognition in social media,the implicit emotion recognition technology based on traditional neural network,tree representation learning and context aware tree representation learning can be studied from this paper.At the beginning of this paper,an implicit sentiment analysis model based on affective goals is proposed,that is,the model which is used in aspect level sentiment analysis task for analysis and research on Chinese implicit sentiment corpus.Firstly,the dataset in the implicit sentiment corpus is preprocessed to get an emotion target word related to the emotion target sentence.Then,based on the characteristics of aspect level affective analysis model,the dataset is uses in different attention modes of the model for comparative experiments.The purpose is to prove the validity of sentiment target words in Chinese implicit affective analysis.The results show that the serialization model Long Short-Term Memory,LSTM is the best in Chinese implicit sentiment dataset.The comparative experiment provides the basic research for the later model.Aiming at the problems of inaccurate feature extraction and gradient explodes or gradient disappears of text information in Chinese implicit sentiment analysis by existing serialization models,a parallel hybrid model of bidirectional long short time neural network and tree Recurrent Neural Network(CA-TRNN)was proposed.The experimental results in the Chinese implicit sentiment analysis task of SMP2019 micro blog show that the model(CA-TRNN)can effectively improve the accuracy of classification results with low training time cost and better application ability.The main work is as follows:1.Aiming at the semantic structure information of implicit sentiment,vanilla tree LSTM is used to initialize the expression of target emotion from this paper,and each node uses recurrent neural network to exchange information with its neighbors repeatedly.This multi-channel information exchange enables each node to obtain more semantic information of sentence level target sentences through rich communication modes.In addition,the number of time steps does not scale with the height of the tree.The target sentence is encoded by a multi hop tree recurrent neural network.2.An attention mechanism coding network is designed to capture the interaction information between hidden features and semantic features between the target sentence and context.The purpose is to introduce the target sentence representation into context aware attention,so that the context representation is trained around the target sentence representation.3.Finally,the semantic representation of the target sentence is obtained by parallel fusion of the semantic information of the context and the target sentence,which can more fully consider the relationship between the context and the target sentence,and overcome the difficulty of Chinese implicit emotion classification is reduced because the target sentence has too long contextFinally,the design and implementation of Chinese implicit sentiment analysis system is designed from this paper.