Practice on a Natural Language Processing use case

--- primary_color: steelblue secondary_color: lightgray text_color: black shuffle_questions: false shuffle_answers: true --- # Why is tokenization strategy crucial for text classification performance? 1. [ ] It determines the computational complexity of the machine learning algorithm 2. [x] It defines how text is split into features, directly affecting what patterns the model can learn 3. [ ] It automatically handles class imbalance in text datasets 4. [ ] It eliminates the need for feature selection in NLP tasks # What is the primary challenge when transferring spam detection models from SMS to email data? 1. [ ] Email messages are always longer than SMS messages 2. [ ] Different programming languages are required for each domain 3. [x] Linguistic patterns, vocabulary, and communication styles differ significantly between domains 4. [ ] Email data requires different machine learning algorithms # In spam detection, why might high precision be more important than high recall in certain scenarios? 1. [ ] High precision reduces computational costs during model training 2. [x] High precision minimizes false positives, preventing important emails from being incorrectly flagged as spam 3. [ ] High precision automatically improves recall in text classification tasks 4. [ ] High precision eliminates the need for cross-validation in model evaluation # What is the main advantage of using CountVectorizer over simple word counting for text feature extraction? 1. [ ] CountVectorizer automatically removes all punctuation from text 2. [ ] CountVectorizer only works with English language text 3. [x] CountVectorizer provides configurable tokenization, vocabulary limits, and preprocessing options 4. [ ] CountVectorizer eliminates the need for train-test splitting # Why is class balancing particularly important in spam detection datasets? 1. [ ] Balanced datasets always achieve higher accuracy scores 2. [x] Imbalanced datasets can lead to models that achieve high accuracy by always predicting the majority class 3. [ ] Class balancing eliminates the need for cross-validation 4. [ ] Balanced datasets require fewer features for optimal performance # What is the primary benefit of stop word removal in bag-of-words text classification? 1. [ ] Stop words always indicate spam content in messages 2. [ ] Removing stop words automatically improves model accuracy 3. [x] Stop words are frequently occurring but often uninformative for classification tasks 4. [ ] Stop word removal reduces the need for tokenization # When building production NLP pipelines, why is modular architecture particularly important? 1. [ ] Modular code automatically achieves better text classification performance 2. [x] Different text preprocessing strategies can be easily tested and optimized independently 3. [ ] Modular architecture eliminates the need for experiment tracking 4. [ ] Modular design automatically handles multiple languages # What is the key difference between accuracy and F1-score in text classification evaluation? 1. [ ] F1-score is always higher than accuracy in spam detection tasks 2. [ ] Accuracy considers only true positives while F1-score considers all predictions 3. [x] F1-score balances precision and recall, while accuracy treats all classes equally 4. [ ] F1-score is only used for multi-class classification problems # Why might a model trained on combined SMS and email data perform differently than domain-specific models? 1. [ ] Combined models always achieve lower performance than single-domain models 2. [x] Mixed training may improve robustness but could dilute domain-specific patterns 3. [ ] Combined datasets require different machine learning algorithms 4. [ ] Mixed training automatically eliminates the need for feature selection # In MLflow tracking for NLP experiments, what is the most important distinction between parameters and metrics? 1. [ ] Parameters track model performance while metrics track configuration settings 2. [x] Parameters are configuration settings (strings/numbers) while metrics are numeric performance measures 3. [ ] Parameters can only be logged once while metrics can be updated during training 4. [ ] Parameters are automatically detected while metrics must be manually specified # What is the primary challenge when applying traditional machine learning evaluation to text classification? 1. [ ] Text data requires specialized machine learning algorithms not available in scikit-learn 2. [ ] Cross-validation cannot be applied to text classification problems 3. [x] Text data often has imbalanced classes and domain-specific patterns requiring careful evaluation strategies 4. [ ] Text classification models cannot be saved or reloaded for deployment # Why is vocabulary size (max_features in CountVectorizer) a critical hyperparameter in text classification? 1. [ ] Larger vocabularies always improve classification performance 2. [ ] Vocabulary size determines which machine learning algorithm to use 3. [x] Vocabulary size affects the balance between feature richness and noise, impacting generalization 4. [ ] Vocabulary size automatically handles class imbalance in the dataset