Deep Learning Based Sentiment Analysis Of Customer Reviews And Its Application For Optimizing Product Design

The prosperity of e-commerce makes a big volume of online customer reviews being generated from time to time.Online reviews contain much valuable information that reflects consumers' attitudes and preferences towards product,which provide new opportunities for product designers understanding customer requirements.In order to help product designers to mine customer requirements from customer reviews,we need to analyze the sentiment in customer reviews.However,most existing approaches to sentiment analysis severely depend on manual-established lexicons and are limited in considering context information,which cause the inaccuracy of sentiment analysis results.In order to overcome the deficiencies of existing approaches,we proposed a deep learning based model of sentiment analysis for customer reviews and optimizing product design.This model is used to help product designers to understand customer requirements and obtain strategies of improving product design.In this model,it starts with crawling online customer reviews from various e-commerce websites,social platform,and critical websites,and pre-processing those data.Then,Word2 Vec is used to train a list of distributed word vectors utilizing well-processed reviews,which are the quantitative representations of customer reviews.Next,short reviews that are more likely to contained product features are extracted from well-processed reviews,and convolutional neural network is then utilized to classify the sentiment polarities of those short reviews.Finally,nouns and noun phrases are extracted as product features candidates,and K-means algorithm is used to cluster those candidates as well as their sentiment labels.After filtering out a little fraction of stop words,the final sentiment analysis results are obtained.Based on the sentiment analysis results,customers' degree of attention and satisfaction towards all product features can be calculated,and the corresponding improving strategies can be drawn.In order to present how to apply the proposed model to recognize customer requirements and obtain the strategies of optimizing product design,we conducted a case study,analyzing online customer reviews of five mobile phones and three computers.On the one hand,we propose the corresponding suggestions for every product.On the other hand,we also found that customers are pervasively concerned with the appearance,the system,the screen,the camera,and the battery of current intelligent mobile phones,and are pervasively concerned with the screen,the appearance,the system,and the keyboard of computers,which can provide references for same kind of product for optimizing product design.The results of case study showed that the proposed model can achieve effective product feature extraction and high accuracy of sentiment classification.It is a reasonable and effective model when it comes to application.