Research On Sentiment Classification By Auto-generated3D Animation Of Chinese Text Messages On Mobile Phones

Full process computer assistant auto-generated animation technology wasintroduced by Acadamician Lu RuQian from CAS. The target of this technology is togenerate animation completely by computer from the beginning of inputtingappropriate story into computer with restricted natural language. In2008,Acadamician Zhang SongMao from CAS applied3D auto-generated technology oncell phone text messages which aimed to send back auto-generated3D animationaccording to sender’s message content.Text sentiment classification is to analyse and classify the emotional inclinationfrom large volume of text messages. It’s a important node in mobile phone3Dauto-generated animation technology, and will benefit following scene planning andexpression of animation information by enriching the emotion information inanimation scene. Current research on text messages sensation classification mainlyfocuses on the duality classification of judgement with less research on multi-labelsensation classification. Most researchers emphasis on phrases and sentances that canexpress emotion inclination in context, i.e. judugement phrases and sentances. Whatdifferent is that we focus on identify Chinese text messages that clearly show thesenders’ sensation by researching on free text messages’ sensation classification. Therewill be no judgement phrases or sentances. There are four dimensions of emotion:happiness, anger, sadness and fear. Research on Chinese sensation inclinationclassification is not so much as on English, and text messages has limited words, lessinformation, free structure and large quantity of vocabulary, so this system is aimingto solve the multi-label classification on text messages’ sensation inclination.Below is the mainly research in this paper:1. Based on HowNet, compute the similarity of two concepts according to thecontext and structure. The compute of two words’ similarity uses the max value ofeach original meaning’s similarity to keep the similarity information among words.2. Applying Simple Bias Model and Support Vector Machine Model to classifyChinese text messages both in objective and subjective. Due to the machine stutymethod, it’s important in gaining the eigenvector. This thesis manually selects fivefeatures as the eigenvector to distinguish subjective Chinese messages and applies theChi-squared test to gain the eigenvector to disginguish objective messages. At the end,combining the two eigenvectors into one eigenvector to classify subjective andobjective Chinese text messages. The test applies TF-IDF to assign weight on featuresof eigenvector.3. On subjective messages, two algorithm ideas are applied in test. One isquestion switch and another is algorithm remaking. In the question switch, Decision Tree and Simple Bias are the basic classifications. In algorithm remaking, MLKNNis applied in test. Eigenvector of Happiness, Anger, Sadness, Fear is also obtainedthrough Chi-squared test. The test applies TF-IDF and Bag-of-Words to assign weighton features of eigenvector.According to the result, Simple Bias Model is better than Support VectorMachine Model in subjective and objective classification of Chinese text messages.The reason is because that there is no obvious separatrix after obtaining features fromChinese messages and mapping to the vector dimension. Decision Tree is better thanMLKNN in subjective multi-label emotion classification of Chinese text messages.Bag-of-Words is best as weight of features.