Skip to content

Latest commit

 

History

History
92 lines (86 loc) · 3.74 KB

README.md

File metadata and controls

92 lines (86 loc) · 3.74 KB

Help Twitter Combat Hate Speech Using NLP and Machine Learning

Objectives

  • Use NLP and ML to make a model to identify hate speech (racist or sexist tweets) on Twitter
  • Use cleaned-up tweets to build a classification model using NLP techniques, cleanup-specific data for all tweets, regularization, and hyperparameter tuning using stratified k-fold and cross validation to get the best model

Prerequisites

  • Sklearn: It includes several effective methods of statistical modeling and machine learning, such as classification, regression, clustering, and dimensionality reduction.
  • Grid search: It is a procedure that thoroughly explores a manually chosen portion of the targeted algorithm's hyperparameter space.
  • Stratified K-fold: This cross validation is an extension of the cross�validation technique used for classification problems.

Dataset Description

Variable - Description
id - Identifier number of the comment
comment_text - The text in the comment
toxic - Status of toxicity with 0 for nontoxic and 1 for toxic

Steps

Task to Perform

  1. Load the tweets file using the read_csv function from the Pandas package
  2. Upload the tweets into a list for easy text cleanup and manipulation
  3. Apply the following steps to clean up the tweets:
  • Normalize the casing
  • Use regular expressions and remove user handles that begin with @
  • Use regular expressions, and remove URLs
  • Use TweetTokenizer from NLTK to tokenize the tweets into individual terms
  • Remove stop words
  • Remove redundant terms like amp and rt
  • Remove # from the tweets while retaining the text that follows it Task to Perform
  1. Use the cleanup process to remove terms with a length of 1
  2. Check the top terms in the tweets:
  • First, get all tokenized terms into one list
  • Use the counter, and find the 10 most common terms
  1. Format the data for predictive modeling:
  • Join the tokens back to form strings, which will be required for the vectorizers
  • Assign x and y
  • Perform train_test_split using sklearn Task to Perform
  1. Use TF-IDF values for the terms as a feature to get into a vector space model
  • Import TF-IDF vectorizer from sklearn
  • Instantiate the model with a maximum of 5000 terms in your vocabulary
  • Fit and apply the vector space model on the train set
  • Apply the model on the test set
  1. Model building: ordinary logistic regression
  • Instantiate logistic regression from sklearn with default parameters
  • Fit model on the train data
  • Make predictions for the train and the test sets Task to Perform
  1. Model evaluation: accuracy, recall, and f_1 score
  • Report the accuracy of the train set
  • Report the recall on the train set: decent, high, or low
  • Get the f_1 score on the train set
  1. Adjust the class imbalance, if any
  • Adjust the appropriate class in the logistic regression model
  1. Train the model again with the adjustment and evaluate
  • Train the model on the train set
  • Evaluate the predictions on the train set: accuracy, recall, and f_1 score Task to Perform
  1. Regularization and hyperparameter tuning:
  • Import GridSearch and StratifiedKFold
  • Choose for C and penalty parameters under the parameters grid
  • Use a balanced class weight while instantiating the logistic regression
  1. Find the parameters with the best recall in cross-validation
  • Choose recall as the metric for scoring
  • Choose a stratified four-fold cross-validation scheme
  • Fit it on the train set
  1. List the best parameters Task to Perform
  2. Predict and evaluate parameters using the best estimator
  • Use the best estimator from the grid search to make predictions on the test set
  • Find the recall on the test set for the toxic comments
  • Find the f_1 score

Setup and Installation:

pip install --upgrade pip
pip install -r requirements.txt
pip list