NLP Sentiment Analysis 4 — Questions and Answers
Question 1: What role does 'transfer learning' play in modern sentiment analysis?
- It transfers sentiment labels from one dataset to another manually
- It leverages pretrained language model weights to improve sentiment classification with less labeled data (Correct answer)
- It moves a trained model from one programming language to another
- It converts audio sentiment to text sentiment
Correct answer: It leverages pretrained language model weights to improve sentiment classification with less labeled data
Transfer learning fine-tunes pretrained models like BERT on labeled sentiment data, achieving high accuracy with fewer examples.
Question 2: Which sentiment analysis technique uses a graph of words to propagate sentiment scores?
- Semantic orientation using PMI
- Graph-based sentiment propagation (Correct answer)
- Naive Bayes with Laplace smoothing
- Gradient boosting on TF-IDF features
Correct answer: Graph-based sentiment propagation
Graph-based propagation spreads known sentiment scores from seed words through a word co-occurrence or similarity graph to label unlabeled words.
Question 3: What is 'emotion detection' and how does it differ from polarity classification?
- Emotion detection classifies text as formal or informal, while polarity detects bias
- Emotion detection identifies specific emotions like joy or anger, while polarity only classifies positive or negative (Correct answer)
- Emotion detection uses audio signals while polarity uses text only
- They are identical tasks with different names
Correct answer: Emotion detection identifies specific emotions like joy or anger, while polarity only classifies positive or negative
Polarity classification assigns positive/negative/neutral labels, while emotion detection maps text to discrete emotion categories such as Ekman's six basic emotions.
Question 4: What is the 'Pointwise Mutual Information' (PMI) method used for in sentiment analysis?
- Measuring sentence perplexity in language models
- Estimating semantic orientation of phrases by comparing co-occurrence with positive vs. negative seed words (Correct answer)
- Computing similarity between two sentiment lexicons
- Evaluating the precision of sentiment classifiers
Correct answer: Estimating semantic orientation of phrases by comparing co-occurrence with positive vs. negative seed words
PMI measures how much more often a phrase co-occurs with positive seed words versus negative ones to estimate its sentiment orientation.
Question 5: Which data augmentation technique is commonly applied to improve sentiment classifier robustness?
- Reducing training set size to prevent overfitting
- Synonym replacement, random insertion, or back-translation to create varied training examples (Correct answer)
- Removing all neutral examples from training data
- Applying L2 regularization to the classifier weights
Correct answer: Synonym replacement, random insertion, or back-translation to create varied training examples
Techniques like synonym replacement and back-translation generate new training sentences that preserve sentiment while varying wording.
Question 6: In attention-based models for sentiment analysis, what does the attention mechanism highlight?
- The most frequent words in the training corpus
- Words most relevant to determining the sentiment of the sentence (Correct answer)
- The grammatical subject of each sentence
- Stop words that should be ignored
Correct answer: Words most relevant to determining the sentiment of the sentence
Attention weights indicate which tokens the model focuses on most when predicting sentiment, often highlighting opinion words and negations.
Question 7: What challenge does 'cross-lingual sentiment analysis' address?
- Analyzing sentiment in code-mixed text within a single language
- Building sentiment models that work across languages, especially for low-resource languages without labeled data (Correct answer)
- Translating sentiment lexicons from English to French only
- Handling typos and informal spellings in English text
Correct answer: Building sentiment models that work across languages, especially for low-resource languages without labeled data
Cross-lingual sentiment analysis aims to transfer sentiment knowledge from high-resource languages (like English) to languages with little or no labeled sentiment data.
NLP Sentiment Analysis 4
What role does 'transfer learning' play in modern sentiment analysis?