In project 2 you used 12 features per tweet and developed a binary classifier. How would you use the same set of features and apply it to a Twitter dataset that has movie ratings with 5 different classes ("Excellent", "Very Good", "Not Bad", "Terrible", "Don't Watch"). Describe the approach you would take to modify your current notebook.
To modify the current notebook for a Twitter dataset with movie ratings having 5 different classes, the approach would involve making several modifications to the existing binary classifier. Below are the steps to be followed:
1. Dataset Preparation:
- Gather a new labeled dataset that contains movie ratings with the five different classes ("Excellent", "Very Good", "Not Bad", "Terrible", "Don't Watch").
- Ensure the dataset is properly formatted with tweet texts and corresponding ratings.
2. Data Preprocessing:
- Apply the necessary preprocessing steps to clean the tweet texts, such as removing HTML tags, URLs, special characters, and punctuation.
- Tokenize the tweet texts into individual words or subwords.
- Convert the words to their base form using stemming or lemmatization.
- Remove stop words that do not carry significant meaning.
3. Feature Extraction:
- Utilize the same set of 12 features previously used in the binary classifier.
- Calculate these features for each tweet in the dataset.
- Normalize or scale the feature values if required.
4. Model Training:
- Modify the model architecture to accommodate the new classification task.
- Update the output layer of the model to have 5 neurons, corresponding to the five different classes.
- Use an appropriate loss function for multi-class classification, such as categorical cross-entropy.
- Train the model on the updated dataset for the new classification task.
5. Model Evaluation:
- Evaluate the performance of the model by computing appropriate metrics like accuracy, precision, recall, and F1-score.
- Generate a confusion matrix to understand the classification performance across different classes.
- Analyze the results and make further improvements if required.
6. Inference:
- Use the trained model for predicting the sentiment of new tweets.
- Preprocess new tweets using the same preprocessing steps as during training.
- Extract the 12 features from the preprocessed tweets.
- Pass the features through the trained model to obtain the predicted movie rating.
By modifying these necessary steps, the existing notebook can be adapted to handle a Twitter dataset with movie ratings having five different classes.