| Online shopping has made great achievements in China for more than ten years,and generated a large number of online product reviews through frequent transactions.On the one hand,it provides a huge database for businesses to analyze consumers' shopping preferences,and provides users with comprehensive shopping decision-making information;on the other hand,it also brings low efficiency of information screening.Comprehensively considering the information characteristics of online product reviews,ranking the effectiveness of product reviews according to users' information adoption preferences,and pushing the most accurate online product information to users will help online trading platforms and businesses to improve users' purchase satisfaction,increase users' trust,and form sustainable competitiveness.At present,the research of online product reviews mainly selects single type of product reviews,focusing on the analysis of the effectiveness of certain review indicators.The screening of influential factors on the effectiveness of online reviews mainly uses subjective research methods such as questionnaire surveys.Machine learning algorithms that efficiently process big data systematically screen the factors that affect the effectiveness of massive online product reviews,this paper improves the index system of online commodity review;secondly,this paper extracts key factors of online product reviews based on machine learning method.In this part,the crawler program is designed to obtain 15169 experiential commodity reviews and 19782 search commodity reviews in Jingdong shopping mall.Multiple regression model,support vector machine recursive feature elimination model and random forest recursive feature elimination model are respectively established to extract the validity indexes of online commodity reviews.The three models are compared based on the standard of root mean square error and model fitting R.The results show that the feature combination under the random forest model is the best,among which there are 5 influencing factors of experiential goods and 7 influencing factors of search goods;Thirdly,forecast of the effectiveness of online product reviews.Taking all influencing factors and key influencing factors of effective online commodity reviews as input variables,the random forest model,support vector machine model,neural network model and logical regression model are selected to classify reviews into "useful" and "useless",and the optimal prediction model is evaluated by precision rate,recall rate and F value.The results show that random forest model based on key factors shows the best prediction effect.In order to further verify,this paper uses the accuracy of random forest model to classify the validity of comments as the auxiliary sorting,and invites users to score the original sorting and the current sorting.The results show that the recommended sorting is recognized by 82.5% of users. |
PDF Full Text Request