Hierarchical Detection Model for Fake Reviews

Abstract

In this study we propose a model to predict fake reviews and fake reviewers Using linguistic features from each review and behavioral features from each reviewer as model features There have been many studies exploring the detection of fake reviews However only few studies have considered both features from review and reviewer And most of the studies use Logistic Model (LM) to predict fake reviews Yet the relationship between reviews and reviewers is nested One reviewer can write more than one review With the data being hierarchical we should use hierarchical analysis instead of using LM for prediction Therefore Hierarchical Logistic Model (HLM) is used as model in this study The dataset is divided into sub-data sets according to different review amount of each reviewer and review time range of the review The experimental results show that HLM can effectively classify fake reviews and fake reviewers due to the hierarchical characteristic of the data In the analysis of fake reviews the use of each reviewer’s reviews within one month can effectively predict fake reviews with an accuracy rate of 86% However it is more effective to predict fake reviewers when we consider more past reviews of each reviewer The best reviewer prediction result is when we consider the reviews within the past 6 years of each reviewer And the accuracy rate is 94% All of the prediction results of HLM are better than other machine learning algorithms such as Support Vector Machine (SVM) Random Forest (RF) Naive Bayes classifier (NB) KNearest Neighbor (KNN)

Date of Award	2020
Original language	English
Supervisor	Sheng-Tun Li (Supervisor)