Hate speech on social media has become a pervasive issue in recent years, causing harm to individuals and communities by perpetuating discriminatory attitudes and even sometimes inciting violence. As such, the ability to detect and classify hate speech in social media is critical in order to confront it.

The objective of this NLP project is to develop the skills necessary to build a model that can effectively detect and classify hate speech in social media, specifically using unstructured data – in this case, labeled Tweets. To achieve this, we will use NLP transformer models to preprocess and tokenize the tweets, applying then various supervised machine learning algorithms to analyze the text and classify it as either hate speech or not.

To evaluate the results, we will be using F1-score on positive cases as success metric, aiming to obtain a result equal or above 0.65

• Transform unstructured data into processable inputs: Text by itself is not a valid numeric input that can be used in ML models.

• Understand the text beyond just formal language rules: The language used on social media is filled with slang and expressions that communicate based on context, making it difficult to process the data with ML models.

• Unbalanced dataset: Just 7.01% of the cases provided in the dataset are positive for hate speech.

The data used for this project is retrived from Kaggle and consists of a CSV file containing 29,530 tweets with the following variables:

• id: (numeric) Tweet ID
• label: (dummy) Tweets that are positive for hate speech are marked with 1, while those that are not are marked with 0.
• tweet: (text) Raw text from the retrived tweets.