Research On Online Reviews Ranking Based On Stacking Method

Affected by the uneven quality of online reviews and the explosive increase in the number of reviews,consumers have begun to gradually change the way they read reviews,from reading a wide range of reviews to only reading the top few or a dozen reviews.In the current actual reviews field,the simple classification ranking of the reviews does not meet the needs of today,and the simple regression ranking of the reviews is not accurate enough and does not meet the current needs of consumers to read reviews.Aiming at the shortcomings of the existing online reviews ranking research in current actual needs,by effectively combining classification ranking and regression ranking,and applying the stacking method to the ranking model,a ranking model based on the stacking method is proposed.This model not only considers the rough problem of ranking results of classification ranking,but also considers the problem of scattered ranking results of regression ranking.In addition,the application of the stacking method is also considered to enhance the effect of the model.The ranking model based on the stacking method is formed by the combination of the classifier based on the stacking method and the regressor based on the stacking method.Online reviews are first classified by the stacking classifier,and the high-quality reviews in the reviews are filtered,and then the high-quality reviews are regressed by the stacking regressor to produce ranking results of high-quality reviews.Experiments on two real online retail product review data sets Amazon Fine Food Reviews and Amazon reviews:Kindle Store Category show that the application of the stacking method in the classifier is AUC,Precision,Recall,and F₁-score,The classification evaluation indicators are superior to the classification base model,and the application of the stacking method in the regressor is superior to the regression base model on the four ranking evaluation indicators NKTD,NDCG@5,NDCG@10,and NDCG@15,the ranking model based on the stacking method is superior to the Light GBM regression ranking model and RF-X-LGB stacking model in the four ranking evaluation indicators.