Research Of Complex Emotion Classification For Microblog Based On Deep Learning

With the social development and technological progress,microblogging and other social media gradually became popular.In daily life,People have become accustomed to using microblog to access information,communicate feelings and express opinions,making it a comprehensive platform for news and entertainment,information dissemination,social interaction and other features.Most of microblog content is text,and mining and analysising the emotions conveyed in this massive microblog text will be able to discove the overall public opinions and sentiment towards hot events,commercial products and public figures,providing real-time and scientific decision-making basis for monitoring public opinion,marketing and public relations crisis.Therefore,the emotion analysis of microblog has a high social significance and commercial value,and has attracted wide attention from researchers.In this dissertation,several deep learning techniques are employed on emotion classification of Chinese microblog,especially for multi-emotion classification,multi-lable emotion classification and multi-lable emotion ranking.The main work includes as follows:(1)This dissertation proposes a Parallel-channel Embedding Covolutional Neural Network(PECNN)for microblog subjective identification and binary classification of positive and negative emotion.While Chinese microblog is hard to segment and error-prone,this dissertation explores traditional machine learning model and Covolutional Neural Network(CNN)model for Chinese microblog emotion classification,using Chinese characters and words as language units have their own advantages respectively.Studies have shown that the convolution operation on word embedding of Chinese characters and words have different emotional semantic composition effect,so this dissertation proposes a Parallel-channel Embedding Covolutional Neural Network(PECNN)to effectively combine both semantic composition together.The experimental results show that the improvements are achived both on microblog subjective identification and binary classification of positive and negative emotion.(2)This dissertation proposes an Emotion-semantics enhanced Multi-channel Convolution Neural Network(EMCNN)for multi-emotion classification of microblog.The importance of emoticon for emotion analysis is explored by hunman annotation of the emotional tendencies.The emotional expression of emoticons and emotional expressions is compared with the emotional tags of word embedding technology and annotated data sets.The MCNN model is used to build the multi-emotion classification of Chinese microblog,and the emotion semantics of MCNN is constructed by using the word embedding of the expression symbol to construct the EMCNN model,and the vector norm preserving algorithm is proposed to simplify the learning process of the model.In the experiment of emotional evaluation task,this model is better than the known best model to achieve better multi-emotion classification performance,for the expression of symbols and expressions without microblog emotional recognition have improved;by organizing more experiments,It is found that this model can use the non-tagged data set and label the data set better,and give full play to the emotional indication of emotional symbols.(3)This dissertation proposes a multi-lable emotion classification framework for sentences in microblog.In the face of multi-label classification problem in microblog emotion evaluation task,traditional text feature representation method based on vector space model is difficult to provide effective semantic features.This paper proposes a multi-label emotion classification framework for microblog sentences.Firstly,we learn the word embedding representation of Chinese words from a large-scale non-annotated microblog text data set,and then use the CNN model to carry out supervised multi-emotion classification learning,The CNN model is used to synthesize the word embedding in the microblog sentence into the sentence vector.Finally,these sentence vectors are used as the characteristic training multi-label classifier to complete the multi-label emotion classification of microblog.(4)This dissertation proposes a deep learning framework for multi-lable emotion classification.The existing depth learning model can not directly deal with multi-label emotional classification problem.This dissertation proposes a deep learning framework for multi-label emotion classification.Combining the depth sorting technique with the representation technology of emotion tag,it can use multi-label emotion model to directly carry out multi-label emotion classification,not only make full use of the emotion provided by multi-label Semantics,but also to learn the vector representation of each emotional tag.