Research On Sentiment Analysis Algorithm Based On Deep Features And Weighted Word2vec Fusion Model

The online evaluation text data contains rich emotional semantic information,which can help consumers understand product information,guide purchase decisions,and provide references for merchants to improve various services.Therefore,accurate mining of emotional semantic information in online evaluation text has great application value.However,existing text analysis methods are difficult to fully mine the emotional semantic features and deep high-level features in the text,which will affect the accuracy of sentiment classification.To this end,this paper proposes corresponding sentiment analysis algorithms for different types of short-text and long-text online evaluation data.The specific research contents are as follows:(1)Existing short text sentiment analysis methods are more difficult to improve the classification accuracy of short text by extracting the emotional and semantic features of the short text at the same time.At present,the method based on sentiment dictionary can extract the sentiment feature of short text more accurately,and the word2vec algorithm can extract the context semantic feature of short text well.Therefore,this paper proposes a sentiment analysis algorithm for short text evaluation based on Sword2vec,which can extract the semantic and emotional features of short text evaluation at the same time.The algorithm first uses the sentiment dictionary method to extract the emotional features of short text,and uses the word2vec algorithm to extract the semantic features of short text;second,uses the emotional features to weight the semantic features to obtain the emotional semantic features of the text Sword2vec;finally,uses Sword2vec feature training The obtained model classifies and recognizes the evaluation text.Experimental results show that the algorithm has improved accuracy and execution efficiency compared with existing traditional methods.(2)Compared with the short-text online evaluation data,the long-text online evaluation sentiment analysis problem is more complicated.Existing long text sentiment analysis methods are more difficult to extract the near and far context semantic information,deep high-level information and sentiment information of long text evaluation data at the same time,and it is difficult to accurately describe the complex features of long text.Therefore,this paper proposes a sentiment analysis algorithm for long text evaluation based on AttBiLSTM_Sword2vec.First,the sentiment value weighted word2vec word vector is used to obtain the sentiment feature and the near semantic feature Sword2vec,and the attention-based bidirectional long-term and short-term memory neural network is used to obtain the depth feature and the far-semantic feature AttBiLSTM of the evaluation corpus;second,the Sword2vec feature and AttBiLSTM The features are fused to get AttBiLSTM_Sword2vec features;finally,the model obtained by AttBiLSTM_Sword2vec feature training is used to classify and recognize the evaluation text.Experimental results show that the algorithm can simultaneously extract and fuse the semantic information,depth information and sentiment information of the near and far context of long text evaluation,thereby improving the effect of sentiment analysis for long text evaluation.