Sentiment Analysis Of Online Reviews Based On Deep Learning

The rise of e-commerce provides a more convenient and efficient platform for enterprises and consumers to make reviews.More and more consumers express their preferences about products through the Internet after shopping,and these comments are strongly subjective.Therefore,how to dig out the emotional information of consumers and potential commercial value in the mass text data has become a hot research spot.Emotional analysis on these review data not only have guiding significance for consumers and enterprises in understanding product information and formulating marketing strategies,but also have reference significance for relevant management institutions.However,traditional sentiment analysis technology can only make emotional judgment at the level of text or sentence,but cannot make the emotional judgment for specific evaluation objects and provide consumers and enterprises with more comprehensive commodity information.Therefore,on the basis of deep learning algorithm,this paper takes real commentary data as the research object,and carries out fine-grained text sentiment analysis from the perspective of specific objectives.Firstly,analyzed the process of sentiment analysis from the coarse-grained and finegrained levels respectively,and the basic theory of deep learning technology,Then expounds the theory of crawler and the collection and preprocessing of comment data,including data cleaning,word segmentation and emotion labeling.Secondly,using Word2 vec technology to transform the words into the word embedding,which avoids the problem of high dimension and irrelevant feature representation,and takes the word embedding generated as the input of the deep learning model,that can improve the accuracy of text sentiment analysis.Finally,we propose an attention based memory network for aspect level sentiment.We taking one or more evaluation objects in the statement as the aspect and using multiple Attention layers to automatically capture the important information about the aspect and using memory network to store the text information,which avoids the problem that loses semantic information due to limited memory for the traditional neural networks.Secondly,proposing the method of double memory module to model the sentences respectively to extract different feature information of sentences.At last,the information extracted by the two modules is combined as the input of classifier.Set up a number of comparative experiments and analyzes the influence of different parameters on the experimental results on the Chinese and English data sets.Compared with other methods,our method can not only express richer semantic features of sentences,but also accurately judge the emotional value of aspect.It's accuracy rate is the highest and It's performance is the best for sentiment classification.