CSE 573: Fake Reviews Detection

Dataset

YelpZip

Statistical machine learning approach

Utilized 3 different machine learning models

Logistic Regression

TF-IDF

Accuracy	Precision	Recall	F1-Score
86.74%	87.13%	99.40%	92.86%

fastText embedding

Accuracy	Precision	Recall	F1-Score
86.34%	86.52%	99.69%	92.64%

SVM

TF-IDF

Accuracy	Precision	Recall	F1-Score
85.65%	85.66%	98.12%	92.25%

fastText embedding

Accuracy	Precision	Recall	F1-Score
86.14%	86.32%	99.63%	92.73%

Random Forest

TF-IDF

Accuracy	Precision	Recall	F1-Score
86.30%	86.40%	99.91%	92.63%

fastText embedding

Accuracy	Precision	Recall	F1-Score
86.35%	86.86%	99.21%	92.62%

Deep learning

• BERT

  BERT base model

Code: BERT

Accuracy	Precision	Recall	F1-Score
86.87%	87.76%	98.62%	92.87%

​

• RoBERTa

RoBERTa base model

Code: RoBERTa

Accuracy	Precision	Recall	F1-Score
86.99%	87.78%	98.74%	92.94%