Research On Emotion Recognition Of User Generated Content Based On Contextual Knowledge

In recent years,with the active development of the Internet and the proliferation of social media,a large number of user generated content has been created.User generated content in emotion recognition can be applied to information consultation,public opinion mining,and maintaining social relationships.How to mine effective emotional information from user generated content to help the active development of advanced artificial intelligence,which has become a hot topic research direction.Based on this,emotion recognition based on dialogue has also received widespread attention from more and more scholars.At present,the emotion recognition method based on deep learning has obtained good research results,and the introduction of the self-attention mechanism has further improved the accuracy of emotion recognition.However,since the self-attention mechanism in the existing self-attention mechanism is relatively common and simple,it is impossible to obtain highquality remote contextual information in dialogue emotion recognition,which affects the improvement of the emotion recognition accuracy.Therefore,this thesis has carried out the following researches:(1)Aiming at the problem that highquality remote contextual information cannot be captured in the user generated dialogue emotion recognition,a context-aware gated recurrent units with selfattention for emotion recognition is proposed.The model introduces the global context into the self-attention mechanism to obtain the overall meaning and syntactic information of words and utterances,thereby enhancing the effect of capturing remote contextual information between words and utterances.Later,the deep context self-attention mechanism is introduced into the model.This method obtains the syntactic information of different types of words and utterances from the deep level,so as to improve the ability of obtaining remote contextual information.(2)Aiming at the problem of semantic sparseness in dialogue content,this thesis adds a convolutional neural network to the emotion recognition model,in which a convolutional layer is used to extract local semantic features,and then a deep-global context self-attention mechanism is introduced.This form of the self-attention mechanism is a combination of global context and deep context,which is mainly used to summarize historical contextual information,so that the model can improve the accuracy of emotion recognition.The experimental results show that the fusion of the global context self-attention mechanism improves the unweighted accuracy by 0.2% and 0.3% respectively compared with the optimal baseline model on the testing dataset of the two public datasets,including Friends dataset and Emotion Push dataset;Furthermore,compared with the global context self-attention mechanism,the introduced the deep context self-attention mechanism improves the unweighted accuracy by 0.5% and 0.2% respectively;In addition,the convolutional neural network based on the deep-global context self-attention is better than the existing latest baseline model in accuracy of partial emotions.