Research On Attribute Extraction Method And Sentiment Analysis Of Innovative Design Products Driven By Online User Reviews

With the booming development of Internet technology and the popularity of online shopping,the number of customers' reviews of products on e-commerce platforms has soared.By analyzing online user comments,the extraction of product attributes and the mining of users' emotional preferences for product attributes provide an important basis for further improving product functions,quality and user satisfaction,supporting the product design process and providing an important basis for designers to make product design decisions.In this paper,a method based on product attribute feature weighting(PAFW)is proposed to extract product review attributes and analyze affective preference intelligently,in order to improve product design quality,shorten Product development and design cycle,and reduce design cost to improve user experience and to improve customer satisfaction and product market competitiveness.This paper proposes a research and analysis method based on product attribute feature weighting(PAFW),which includes the following aspects: firstly,it proposes a TFIDF-Kano(TIK)model based on word frequency inverse document to automatically recognize and extract product attribute keywords.Then,the improved deep learning model(TIK-PAFW)and various deep learning models were used to carry out comparative experiments and results analysis for product attribute extraction based on the improvement of BI-LSTM-CRF.Finally,based on the proposed PAFW method,the weighted method(CTI)is introduced into a variety of deep learning models to carry out comparative experiments and result analysis of product review sentiment analysis with the original model.The main research contents of this paper are as follows:1.Aiming at the problems of high labor cost and low extraction efficiency in traditional manual or semi-automatic text keyword extraction,this paper proposes an automatic extraction method of keywords related to product attributes based on TIK model.Firstly,the words with higher frequency in the document are calculated by the word frequency statistics in the inverse document frequency method.Then the document is divided into several subdocuments,and the inverse document frequency value is calculated to avoid the influence of common words on keywords,and then the inverse document frequency value of the word frequency of each word.Then,kano model is used to analyze the correlation of product attributes and select the keywords of product attributes.Finally,through experimental verification,the TIK model of keyword extraction is an innovative exploration and method,which can accurately and efficiently realize the automatic extraction of keywords.2.Aiming at the problems of low efficiency and high cost of extracting important information from text by manual analysis of product reviews,this paper proposed to improve the BI-LSTM-CRF model based on PAFW method and combined with the TIK model of product attribute extraction,and constructed a weighted short and long memory neural network(TIK-PAFW)model.By comparing with HMM,CRF,LSTM and BI-LSTM,it is verified that the improved TIK-PAFW model has higher extraction accuracy of product attribute keywords,recall rate and F1 in the self-built e-commerce review data set taking mobile phones as an example.The accuracy rate,recall rate and F1 rate of the weighted long and short memory neural network model were 93.89%,94.52% and 94.60%,which reached the highest among several models.Finally,through the experimental results,it is concluded that the TIK-PAFW model based on the PAFW experimental method has a higher extraction accuracy,recall rate and F1 score for the extraction of product attribute keywords,which verifies the effectiveness of the PAFW method.3.In view of the complex process of traditional text sentiment analysis based on sentiment dictionary and the high difficulty of establishing sentiment dictionary,this paper proposes a deep learning method based on PAFW to build a weighted neural network model to conduct sentiment preference analysis on product reviews.Firstly,the process of product attribute sentiment analysis is designed,a weighted model of feature selection and feature weighting(CTI)is proposed,and CTI-CNN,CTI-LSTM and CTI-BI-LSTM models are constructed to carry out comparative experiments with the existing three models in the self-built product corpus,so as to realize the analysis of product attribute sentiment preference.Then,through comparison and verification experiments,it is concluded that the extraction accuracy of the CTI-based model is higher than that of the traditional deep learning model.The accuracy of various deep learning models can reach 96.99% at the lowest level,while the extraction accuracy of the improved CTI-BI-LSTM model reaches 98.08%,and the loss value also converging to 0.0734.Finally,it is demonstrated that the deep learning method based on PAFW proposed in this paper has a strong performance of sentiment analysis of product reviews.