Research On Comment Sentiment Analysis Based On Deep Learning Algorithm

With the advancement of Internet technology,e-commerce is booming,and more and more commodity transactions take place in the Internet.Merchants sell their products on the Internet,and users are happy to share their opinions and experiences about the products after purchasing them online.As a result,product information and comment text in the network quickly expanded.If it is difficult to collect and process massive amounts of information on the Internet by means of manual methods,text analysis is needed to help users quickly obtain important information from hundreds of millions of comments,and sentiment analysis technology emerges as the times require.The construction of a high-quality dictionary based on the lexicographic method requires a lot of manpower.The machine learning method relies too much on the characteristics of the sentence vector.These features come from the manual selection,which leads to different analysis results.As a self-learning classification method,the deep learning method can achieve better results on the sentiment analysis task without excessive manual intervention.Therefore,the study of natural language analysis based on deep learning has great theoretical and practical value.Aiming at the low accuracy rate of traditional text sentiment analysis methods and the low efficiency of deep learning methods in training,testing and analysis,this paper deeply studies the deep emotion-based commentary sentiment analysis technology and classifies the efficient texts of principal component analysis(PCA).The Fasttext method is used as a text vector generation algorithm to improve the quality of the generated text vector.Combining the gated cyclic neural network(GRU)with the convolutional neural network(CNN),the Attention-CNN-GRU comment text sentiment analysis model is established to improve the correct rate of sentiment analysis results and model training efficiency.The specific research content is as follows.In order to obtain better accuracy and shorter training test time,the Attention-CNN-GRU model combined with convolutional neural network and gated cyclic neural network is proposed.The neurons in the traditional neural network are all connected,and there is no connection between the neurons.The sample processing is independent of each other,so the time series changes cannot be processed.A gated loop neural network can process sentences using timing relationships,store historical context information,and can take into account subsequent context information.The gated circulatory neural unit combines the forgetting and memory windows on the basis of the long-and short-term memory neural unit,and only consists of the update gate and the reset gate,which reduces the internal calculation amount of the unit,and the calculation efficiency is greatly improved.And the importance degree of different words for different tasks is different.On the basis of CNN-GRU model,attention mechanism is added to obtain Attention-CNN-GRU model.The role of the convolutional neural network is to continually train the hidden features in the comment text and combine them to achieve the purpose of feature learning selection.Aiming at the problem of the influence of key parameters on the training test results in the deep learning network,the experiment compares the effects of the three parameters of learning rate,Dropout and Batchsize on the results.The experiment can be seen that the learning rate follows the law that the learning rate is smaller and the learning rate is smaller.The selection of the abstaining coefficient is to determine the appropriate amount.If the model is too large,the effect of the model will be reduced.The batch size also needs to be combined with the effect of the model.The experiment proves that the model has a high accuracy rate and better time efficiency in the evaluation of emotional classification tasks,and has theoretical and practical value for the analysis of emotional analysis.