Hate Speech Detection Model

The term hate speech is understood as any type of verbal, written or behavioural communication that attacks or uses derogatory or discriminatory language against a person or group based on what they are, in other words, based on their religion, ethnicity, nationality, race, colour, ancestry, sex or another identity factor. In this article, I will take you through a hate speech detection model with Machine Learning and Python.

Hate Speech Detection is generally a task of sentiment classification. So for training, a model that can classify hate speech from a certain piece of text can be achieved by training it on a data that is generally used to classify sentiments. So for the task of hate speech detection model, I will use the Twitter data.

Also, Read – Linear Regression with PyTorch.

Hate Speech Detection Model

The data set I will use for the hate speech detection model consists of a test and train set. The training package includes a list of 31,962 tweets, a corresponding ID and a tag 0 or 1 for each tweet. The particular sentiment we need to detect in this dataset is whether or not the tweet is based on hate speech. You can download the dataset from here.

So, let’s get started with the task of building a hate speech detection model. I will simply start with reading the datasets by using the pandas package in python:

Training Set: (31962, 3) 31962
Test Set: (17197, 2) 17197

Data Cleaning

Data cleaning is the process of preparing incorrectly formated data for analysis by deleting or modifying the incorrectly formatted data which is generally not necessary or useful for data analysis, as it can hinder the process or provide inaccurate results. Now I will perform the process of data cleaning by using the re library in Python:

Handling Imbalanced data for Hate Speech Detection Model

If you will deeply analyse the task we are working on with context to the data we are using, you will find that the tweets regarding hate speeches are comparatively lesser than others, so this is a situation of an unbalanced data.

If we will fit this data to train our hate speech detection model, then the model will not generalize any hate speech because the data with context to the hate speech is very less than the positive ones. So in this situation, we need to prepare the data to fit properly in our model.

There are a number of methods you can use to deal with this. One approach is to use either oversampling or downsampling. In the case of oversampling, we use a function that repeatedly samples, with replacement, from the minority class until the class is the same size as the majority. Let’s see how we can handle this:

1 29720
0 29720
Name: label, dtype: int64

Creating a Pipeline

For simplicity and reproducibility of the hate speech detection model, I will use the Scikit-Learn’s pipeline with an SGDClassifier, before training our model:

Training the Hate Speech Detection Model

Now, before training the model, let’s split the data into a training set and a test set:

Now let’s train the model and predict the results on the test set using the F1 score method:


Also, Read – Telegram Bot with Python.

So we got an F1 score of 0.96 per cent which is generally appreciatable. This model can now be deployed and used in production. I hope you liked this article on Hate Speech Detection Model with Machine Learning. Feel free to ask your valuable questions in the comments section below. You can also follow me on Medium to learn every topic of Machine Learning. 

Follow Us:

Leave a Reply