Text Classification Flashcards

What is the primary function of a softmax layer at the output of a text classification neural network?

Answer: To convert raw logits into a probability distribution over all classes

Softmax exponentiates each logit and divides by the sum, producing a valid probability distribution where all class probabilities sum to 1.

In the context of text classification, what is 'data augmentation'?

Answer: Artificially expanding the training dataset by creating modified versions of existing examples

Text data augmentation techniques such as synonym replacement, back-translation, and random insertion create new training examples to improve model robustness and generalization.

Which approach best describes 'prompt-based' text classification using large language models?

Answer: Framing the classification task as a text generation problem with natural language instructions

Prompt-based classification provides the LLM with a natural language template (prompt) so the model generates or selects the correct label token, unifying classification with the generative pretraining objective.

What distinguishes 'domain adaptation' in text classification?

Answer: Adapting a model trained on one domain (e.g., news) to perform well on a different domain (e.g., medical)

Domain adaptation addresses the distribution shift between source and target domains, using techniques like continued pre-training or adversarial training to transfer knowledge effectively.

What is 'macro-averaged F1-score' in multi-class text classification?

Answer: The unweighted mean of the per-class F1-scores, treating all classes equally

Macro-averaged F1 computes F1 independently for each class and then takes the simple average, giving equal weight to each class regardless of its frequency.

Which of the following is a key characteristic of convolutional neural networks (CNNs) when applied to text classification?

Answer: They apply filters over local n-gram windows to capture local features

CNNs for text use 1D convolutional filters that slide over windows of consecutive words, learning to detect locally informative n-gram patterns useful for classification.

What is the role of 'attention pooling' in a text classification model?

Answer: It learns to assign higher weights to more task-relevant tokens when summarizing a sequence

Attention pooling computes a weighted sum of token representations where weights are learned based on relevance to the classification task, producing a more informative document-level embedding.