| With the rapid development and increasing openness of the Internet,the emotional classification of subjective content on the Internet has become a hot spot for text information processing.Big Data research is becoming more and more popular,people have more needs for data analysis,the liberalization and diversification of the text expression cause more complex sentence patterns in the text and affect the results of sentence-level sentiment analysis.At present,the traditional affective classification method is not efficient for the emotional analysis of complex sentence patterns.According to the structural characteristics of complex sentences and the effective analysis of complex sentences.Firstly,this study made a detailed emotional analysis of various complex sentences.Summing up the seven complex sentence patterns,dividing conjunctions into three categories,according to the classification of complex sentences,summing up complex sentences model,an emotional classification method for complex sentences is proposed.The establishment and expansion of the emotional dictionary is the basis for the research work of text emotion analysis and the perfection of theemotional dictionary helps to improve the accuracy of emotional analysis.Secondly,this study processing the extension of emotional lexicon based on Word2 vec,the emotional words in the basic emotional dictionary are used as seed words,by using the word embedding model trained by Word2 vec,to obtain the words with larger cosine values of the seed emotion words,get the synonyms for the seed emotional words,then used improved How-net vocabulary semantic similarity calculation method to filter the emotional words produced by the previous step,and collecting network hotspot words,completing the construction of an emotional dictionary.This study divided emotional tendencies into three categories: positive,negative and neutral.Thirdly,building a complex sentence pattern CSSCM and combining with SVM for training and prediction,then classify complex sentence and random sentence separately.Finally,in this paper,a multichannel variable filter dynamic muli-pooling convolution neural network model MVDCNN is proposed for emotional classification,in the convolution layer extracting feature vectors of different granularity words,combined with different word embedding(Word2vec+Glove)pretraining word embedding.The dynamic multi-pooling method segmented complex sentences,extraction maximum of features of complex sentences,completing the three classifications of emotion through training model.In the final experiment,comparing the emotional classificationaccuracy rate of SVM,CNN,CSSCM and MVDCNN models.Based on SVM,CNN,CSSCM,MVDCNN models sentiment classification accuracy in random sentences:50.6%,52.0%,53.7%,55.2%,accuracy in complex sentences:51.5%,52.8%,54.9%,56.3%.The experimental results show that the new complex sentence sentiment classification model CSSCM has a significant improvement in the processing accuracy compared with the traditional sentiment classification method,the MVDCNN model further improves the accuracy rate of the sentiment classification,and get good results of emotional analysis.The experimental results show that the method proposed in this study makes the accuracy rate of sentiment classification of the complex sentences improved to different degree.This study further improves the performance of the system and shows excellent performance. |
