Research On Sentiment Polarity Analysis Of E-commerce Reviews Based On Deep Neural Network

With the rapid development of Internet technology,e-commerce has been fully integrated into people's daily lives.When consumers cannot directly understand products,reviews play a key role.There is a lot of valuable information hidden in product reviews.Mining the emotional tendencies of these reviews can not only allow consumers to indirectly understand the quality of the product,but also allow merchants to adjust their strategies in a timely manner through consumer feedback.However,in the previous research on sentiment analysis of e-commerce reviews,there are some shortcomings.First,when using traditional matrix filling algorithms to convert e-commerce reviews into a fixed-size text vector matrix,the text vector matrix takes up a lot of storage space and the model runs for too long.Second,the model is less robust.Models often have good results in certain e-commerce comment,but they are unstable in other e-commerce comments.Finally,in the past traditional machine learning models and deep learning network models,the accuracy,precision,recall,and F1-Measure values of sentiment analysis for e-commerce reviews are not ideal.In view of the above shortcomings,the main research work of this paper is as follows:First,this article made improvements on the word vector training corpus,combining the Wikipedia Chinese corpus and product review corpus as a mixed corpus for word vector training to generate a word vector library.Compared with using only the Wikipedia Chinese corpus,the experimental results show that the robustness and accuracy of sentiment classification of the five kinds of e-commerce reviews under the mixed corpus are significantly improved.Secondly,in order to solve the problem that the storage space of the text vector matrix is too large and the running time of the deep neural network model is too long,this paper uses the characteristics of the Gaussian distribution to propose two new matrix filling algorithms,namely the standard length zero matrix filling algorithm(ZMFtoSL)And standard length cyclic matrix filling algorithm(CMFtoSL),then use deep learning neural network to analyze the sentiment polarity of five kinds of e-commerce reviews to prove the effectiveness of two new filling algorithms.The experimental results show that compared with the traditional matrix filling algorithm,the two algorithms proposed in this paper perform well in the space occupied by the text vector matrix,the model running time,and the accuracy rate.Finally,this paper proposes a MCNN-LSTM neural network model based on the soft attention mechanism,which combines a multi-scale convolutional neural network(MCNN)with a long short-term memory network(LSTM)and lead into the soft attention mechanism.The experiments show that compared with the traditional machine learning models,CNN and CNN-LSTM deep learning models,the deep neural network model proposed in this paper improves the average accuracy rate of five kinds of e-commerce reviews by 3.34 percentage points,and has better Robustness.The model's accuracy,recall,and F1-Measure values have improved significantly.