ankit-bhagat851 / hate-speech-detection

Hate-Speech-Detection

We studied the theoretical concept of hate speech. Its definition, laws and classification. Second, we collected dataset form online platforms as labelled and unlabeled data. Third we preprocessed it and analysed the dataset. Fourth we implemented 10 machine learning algorithms namely Logistic Regression, Decision Tree Classifier, Gradient Boosting Classifier, Random Forest Classifier, AdaBoost Classifier, XGB Classifier, Gaussian Naive Bayes, Multinomial Naive Bayes, Stochastic Gradient Descent Classifier, KNeighbour Classifier, Support Vector Machine. Finally, we extracted a dataset from twitter of nearly 50,000 tweets and then performed the above mentioned algorithms.

Result of 1 lakh labelled tweets

Name	Accuracy BOW	Accuracy TF-IDF
Logistic Regression	0.79	0.79
Decision Tree Classifier	0.74	0.75
Gradient Boosting Classifier	0.76	0.76
Random Forest Classifier	0.80	0.80
Adaboost Classifier	0.76	0.76
XBG Classifier	0.76	0.76
Gaussian NB	0.70	0.73
Multinomial NB	0.72	0.74
SGD Classifier	0.78	0.78
K Neighbours Classifier	0.74	0.72

Result of 50,000 unlabelled tweets

Name	Count of Hate speech BOW	Count of Not Hate speech BOW	Count of Hate speech TF-IDF	Count of Not Hate Speech TF-IDF
Logistic Regression	40710	10408	41182	9936
Decision Tree Classifier	27677	23441	28408	22710
Gradient Boosting Classifier	50301	817	50211	907
Random Forest Classifier	31420	19698	32731	18387
Adaboost Classifier	48800	2318	48367	2751
XBG Classifier	50463	655	50299	819
Gaussian NB	19664	31454	25992	25126
Multinomial NB	34250	16868	37242	13876
SGD Classifier	48308	2810	46974	4144
K Neighbours Classifier	29411	21707	26241	24877