Emotional Analysis Of Internet Media Based On Fine Granularity

With the popularization of the Internet,more and more people choose to express their views and opinions on the network media.The content involved has become an indispensable source of information in daily life and work.How to further enhance the computer in processing,understanding and use of human language contains emotional information has become a hot issue of concern.Compared with paying attention to the overall emotion of the text,the comments of the network media involve more complicated details.Therefore,it is necessary to mine and analyze the emotional tendency of online media comments from a fine-grained perspective.However,there are still some problems in fine grained affective analysis,such as limited ability to express emotion in complex context,difficult to capture the relevance of long text semantics,poor generalization ability of the model,low classification accuracy,etc.Based on the above problems,this paper proposes a fine-grained network media affective analysis model based on the ERNIE(Enhanced Representation through Knowledge Integration)and the DPCNN(Deep Pyramid Convolutional Neural Networks).First,the local semantic information of input samples and the dependence of long text cannot be well captured by the BERT pre-training language model.ERNIE preprocessing language model is used in this model.By adding more high-quality corpus and fusing more external knowledge information,it can solve the problem of long-distance Chinese semantics.Second,in text categorization,the Text CNN network cannot capture semantic relationships between long-distance texts.Through the construction of DPCNN network,the depth of the network is deepened,and the correlative semantics of long-distance texts can be extracted.The effect of the fine-grained affective analysis model is compared with the experimental accuracy and loss function.Experimental results show that the proposed ERNIE-DPCNN model is more accurate and the test results are more in line with the true emotion expressed by users.