Sentiment Analysis Of Chinese Short Messages Based On Deep Learning Of Mobile Animation

With the development of artificial intelligence and computer-related technologies,in the 1990 s,Academician Lu Ruqian of the Institute of Mathematics of the Chinese Academy of Sciences proposed the automatic generation technology of the whole process computer-assisted animation combined with artificial intelligence and computer animation technology.In recent years,the mobile network developed rapidly.In 2008,the researcher of the Chinese Academy of Sciences Zhang Songmao proposed the idea of applying the whole process of computer-assisted animation generation technology to the SMS of mobile phones and completed the design and implementation of a mobile phone 3D animation automatic generation system.The system combines artificial intelligence technology,and computer animation related technologies.For the inputting Chinese phone text messages,after the four steps of information extraction,qualitative plot planning,quantitative animation calculation,and network rendering,the generated video animations are finally sent to the short message receiver.Information extraction is the first step of the mobile phone 3D animation automatic generation system,which is located in a key position.Its main role is to extract the animated information in the Chinese SMS text,and the message sentiment analysis is used as part of the information extraction to analyze the sentiment in the short message.The sentiment information contained in it provides a strong reference for the expression of emotions in the follow-up plot planning.The accuracy of the existing sentiment analysis system in the animation system is not high,and the practical purpose cannot be achieved.For this reason,this paper proposes using deep learning methods for the sentiment analysis of Chinese text messages and classifies the sentiment of text messages into four categories: happy,anger,sad,and fear for the expressiveness of the animation.The research work of this paper mainly includes the following two parts.First,we designed and implemented a sentiment analysis model LSTM_CNN combining Long Short-Term Memory(LSTM)and Convolutional Neural Network(CNN)to analyze the sentiment of text messages in Chinese.Among them,text uses word vector technology to represent,the convolutional neural network model uses convolution and pooling operations for feature extraction,the LSTM model preserves the sequence information of the text,uses its unique gate structure for feature extraction.Our experiment uses SMS-2008 to annotate more than 100,000 short messages in the Chinese short message database as the initial corpus.After pre-processing,18,163 short messages containing happy,anger,sad and fear emotions are obtained and we train the classification model on this data set.The experimental results show that the performance of the LSTM_CNN model combined with the long-short memory network and the convolutional neural network we designed is better than that based on only convolutional neural network and only long-short-term memory network and the accuracy rate is 84.45% and the recall rate is 84.56%.Second,we designed and implemented a topic-based attention convolutional neural network sentiment analysis model.The attention mechanism can automatically calculate the importance of each word in the text message to the target output result,namely the attention weight.The topic is the conciseness of the content of the short message and is the central idea of the short message.The information extraction module in the mobile 3D animation system analyzes the topic of the short message through a rule-based method.In order to fuse the attention mechanism with the information extraction in the mobile 3D animation system,this article innovatively proposes to extract the topic as part of the attention mechanism and combine it with the convolutional neural network to analyze the sentiment of short message text.After training,the accuracy of the classification model reached 85.82%,indicating the effectiveness of the introduction of topic and attention mechanism methods.Finally,the mobile phone 3D animation generation system chooses to use this model for sentiment analysis.During the operation of the mobile phone 3D animation automatic generation system from December 7,2017 to March 20,2018,we conducted an open experiment on the subject-based attention-based convolutional neural network classification model.The system received a total of 230 SMS messages,of which,due to system update questions and so on 30 messages were error,and 136 short messages were classified as objective.For the subjective 64 messages,our sentiment analysis system was correctly classified to 54 and the accuracy rate was 84.38%.This achieves practical purposes in the mobile phone 3D animation automatic generation system.This paper designs and implements a Chinese short message sentiment analysis system based on the topical attention convolutional neural network model for the automatic generation of mobile 3D animation.Open experiments show the effectiveness of the model.Further work includes integrating multiple models and processing unbalanced corpora,processing polysemy,etc.to improve performance.