NLP Sentiment Analysis 2 — Questions and Answers
Question 1: Which challenge does 'negation handling' address in sentiment analysis?
- Detecting sarcasm in text
- Reversing sentiment when words like 'not' or 'never' appear (Correct answer)
- Handling multilingual text
- Identifying named entities
Correct answer: Reversing sentiment when words like 'not' or 'never' appear
Negation handling ensures that words like 'not' flip the sentiment of subsequent words (e.g., 'not good' becomes negative).
Question 2: What is 'aspect-based sentiment analysis' (ABSA)?
- Analyzing sentiment of entire documents
- Identifying sentiment toward specific aspects or features within a text (Correct answer)
- Classifying text by topic before sentiment analysis
- Measuring sentiment intensity on a scale
Correct answer: Identifying sentiment toward specific aspects or features within a text
ABSA extracts sentiment at the level of specific aspects (e.g., 'battery life' or 'camera') rather than the whole review.
Question 3: Which metric is most appropriate for evaluating sentiment analysis on a highly imbalanced dataset?
- Accuracy
- F1-score (Correct answer)
- Perplexity
- BLEU score
Correct answer: F1-score
F1-score balances precision and recall, making it more informative than accuracy when class distribution is skewed.
Question 4: What does the term 'valence' refer to in sentiment analysis?
- The grammatical structure of a sentence
- The positive or negative direction of an emotional expression (Correct answer)
- The frequency of sentiment words in a corpus
- The confidence score of a classifier
Correct answer: The positive or negative direction of an emotional expression
Valence captures whether an expression is positive or negative, forming the core dimension of sentiment polarity.
Question 5: Which approach is used in 'distant supervision' for sentiment analysis?
- Using human-annotated labels for training
- Automatically labeling data using heuristics such as review star ratings (Correct answer)
- Training on synthetic data generated by a language model
- Transferring labels from a related language
Correct answer: Automatically labeling data using heuristics such as review star ratings
Distant supervision exploits noisy but abundant labels (like star ratings) to train sentiment classifiers without manual annotation.
Question 6: What problem does 'domain adaptation' solve in sentiment analysis?
- Handling multiple languages in a single model
- Improving model performance when training and test data come from different domains (Correct answer)
- Reducing model size for deployment
- Detecting irony in product reviews
Correct answer: Improving model performance when training and test data come from different domains
Domain adaptation addresses the drop in performance when a model trained on one domain (e.g., movies) is applied to another (e.g., electronics).
Question 7: In the context of sentiment lexicons, what is 'SentiWordNet'?
- A neural network architecture for sentiment classification
- A lexical resource that assigns positive, negative, and objective scores to WordNet synsets (Correct answer)
- A dataset of annotated product reviews
- A tool for detecting sarcasm in social media
Correct answer: A lexical resource that assigns positive, negative, and objective scores to WordNet synsets
SentiWordNet extends WordNet by assigning sentiment scores (positivity, negativity, objectivity) to each synset.
NLP Sentiment Analysis 2
Which challenge does 'negation handling' address in sentiment analysis?