Profiling reviewers’ social network strength and predicting the “Helpfulness” of online customer reviews

Highlights：

• The Social Network Strength (SNS) features of reviewers are introduced.

• A comparison of the regression and classification approaches used to predict the helpfulness of the reviews is provided.

• A dataset of 90,671 Yelp shopping reviews is used for evaluation.

• Using Bagging Gradient-Boosted Trees (B-GBT), all features and datasets of recent reviews provide the best performance for both tasks.

• The Random Forest (RandF) weights show that the key features belong not only to the review features, but also to the business and reviewer features.

摘要

•The Social Network Strength (SNS) features of reviewers are introduced.•A comparison of the regression and classification approaches used to predict the helpfulness of the reviews is provided.•A dataset of 90,671 Yelp shopping reviews is used for evaluation.•Using Bagging Gradient-Boosted Trees (B-GBT), all features and datasets of recent reviews provide the best performance for both tasks.•The Random Forest (RandF) weights show that the key features belong not only to the review features, but also to the business and reviewer features.

论文关键词：Online reviews,Review helpfulness,Information overload,Predictive modeling,User profiling,Feature engineering,Feature selection

论文评审过程：Received 6 July 2020, Revised 10 November 2020, Accepted 14 December 2020, Available online 18 December 2020, Version of Record 8 January 2021.