Results

Results

Here we will keep updated the current best results of our research in review deception detection, using different algorithms, parameter sets, and approaches. All of the work that produces these results can be found in the /notebooks directory in the LUCAS subdirectory of the project source code.

Statistical Models

Classifier	Dataset	Accuracy
Log. Regression	OpSpam	~0.87
Naive Bayes	OpSpam	~0.87
Linear SVC	OpSpam	~0.88
k-NN	OpSpam	~0.8
Log. Regression	YelpData	~0.72
Naive Bayes	YelpData	~0.67
Linear SVC	YelpData	~0.75
k-NN	YelpData	~0.59
LDA	YelpData	~0.68

Neural Models

Classifier	Dataset	Accuracy
FFNN(BOW)	OpSpam	~0.86
FFNN(OpSpam W2V)	OpSpam	~0.57
FFNN(Pretrained W2V)	OpSpam	~0.66
FFNN(Pretrained W2V)	Yelp	~0.62
CNN(BOW)	OpSpam	~0.82
CNN(OpSpam W2V)	OpSpam	~0.75
CNN(Pretrained W2V)	OpSpam	~0.78
CNN(Pretrained W2V)	Yelp	~0.71
LSTM (BOW)	OpSpam	~0.88
LSTM (OpSpam W2V)	OpSpam	~0.65
LSTM (Pretrained W2V)	OpSpam	~0.65
LSTM (Pretrained W2V)	Yelp	~0.70
BiLSTM (Pretrained W2V)	Yelp	0.65
FFNN(BOW)	YelpData(:20000)	~0.64
FFNN(Yelp W2V)	YelpData(:20000)	~0.63
BERT	YelpData(:10000)	0.65