Classification for NLP

A model achieves high accuracy on a test dataset, but a deeper analysis shows it performs poorly on the minority class. What does this indicate, and how should you evaluate the model instead?

1. 

• The model is overfitting. So, use cross-validation

• The model has high recall. So, evaluate using a confusion matrix

• The dataset is imbalanced. Use metrics such as F1 Score and ROC AUC to evaluate

• The model is underfitting. Try a different algorithm

You’re working on a disease detection model where missing a positive case could have severe consequences. Which evaluation metric should you prioritize and why?

• Precision, because false positives must be minimized

• Recall, because false negatives must be minimized.

• F1-score, because it balances precision and recall.

• Accuracy, because it reflects overall model performance.

Two models produce identical accuracy scores but have different confusion matrices. What can you infer from this?

• Both models perform equally well.

• Both models have the same recall and precision.

• The models handle false positives and false negatives differently.

• All of the above

When is the F1-score a better evaluation metric than accuracy?

• When the test data is balanced.

• When both, Precision and Recall are important.

• When you want to optimize for true negatives.

• When the model has high AUC-ROC.

How does the ROC curve help in comparing models?

• It shows the model’s accuracy at different thresholds.

• It visualizes the trade-off between true positive rate and false positive rate.

• It compares the number of true positives and true negatives.

• It measures the correlation between actual and predicted probabilities.

A spam detection system is tested using precision and recall. Precision is very high, but recall is low. What does this indicate about the model’s behavior?

• The model predicts fewer spam emails but is highly confident in its predictions.

• The model predicts most emails as spam, including false positives.

• The model is balanced and performs well overall.

• The model has a high rate of false positives and false negatives.

What makes text classification different from other types of classification tasks?

• It may require special algorithms for complicated classification tasks

• The input data is categorical rather than numerical.

• The input data is unstructured and requires preprocessing like tokenization and embedding.

• Both A and C

• All of the above

What are the advantages of using word embeddings from BERT in text classification tasks?

• To convert categorical data into numerical data

• To improve the accuracy of the model

• To capture semantic relationships between words in a high-dimensional space.

• To increase the speed when using deep learning algorithms.

• C and D

Which of the following algorithms produce word embeddings?

• Word2Vec

• Glove

• Tfidf Transformer

• Bag of words

• All of the above

A sentiment analysis model trained on product reviews performs poorly on medical text. What is most likely the issue?

• The training data is not large enough.

• Sentiment analysis cannot be applied to medical text.

• The vocabulary and context of the training data do not match the new domain.

• The model’s architecture is unsuitable for classification tasks.