NLP Sentiment Analysis 5 — Questions and Answers
Question 1: What is the 'semantic orientation' of a word in sentiment analysis?
- The grammatical case assigned to a noun
- The degree to which a word is associated with positive or negative sentiment (Correct answer)
- The part of speech tag assigned to a word
- The topic category a word belongs to
Correct answer: The degree to which a word is associated with positive or negative sentiment
Semantic orientation quantifies how strongly a word skews positive or negative based on its associations and usage patterns.
Question 2: Which issue arises with 'word-level' sentiment analysis that 'sentence-level' analysis helps mitigate?
- Ignoring document structure
- Missing context that determines whether a word is used positively or negatively in the sentence (Correct answer)
- Being too slow for real-time applications
- Requiring too much training data
Correct answer: Missing context that determines whether a word is used positively or negatively in the sentence
At word level, 'unpredictable' looks negative, but sentence context like 'unpredictably funny' reveals a positive sentiment.
Question 3: What is 'sentiment composition' in the context of phrase-level analysis?
- Combining outputs from multiple sentiment classifiers
- Computing the sentiment of a phrase by composing sentiments of its constituent words using linguistic rules (Correct answer)
- Segmenting a document into sentiment-bearing phrases
- Creating training datasets by composing existing examples
Correct answer: Computing the sentiment of a phrase by composing sentiments of its constituent words using linguistic rules
Sentiment composition uses rules (e.g., modifier + adjective) to derive phrase-level sentiment from word-level scores and syntactic structure.
Question 4: In the context of Twitter/social media sentiment analysis, what makes the task particularly challenging?
- Excessive formality and long sentence length
- Informal language, abbreviations, emojis, sarcasm, and very short text with little context (Correct answer)
- Strict grammar rules that confuse classifiers
- Absence of named entities
Correct answer: Informal language, abbreviations, emojis, sarcasm, and very short text with little context
Social media text is noisy with abbreviations, hashtags, emojis, and sarcasm, all of which break assumptions of standard NLP models.
Question 5: What does a 'Recursive Neural Network' (RecNN) exploit for sentiment analysis?
- The sequential order of words in a sentence
- The parse tree (syntactic structure) of a sentence to compose sentiment bottom-up (Correct answer)
- The co-occurrence statistics of words in a corpus
- The document-level topic distribution
Correct answer: The parse tree (syntactic structure) of a sentence to compose sentiment bottom-up
RecNNs apply composition functions at each node of a parse tree, building phrase-level sentiment representations recursively from leaf words.
Question 6: Which evaluation benchmark is widely used for fine-grained sentiment analysis of movie reviews?
- SQuAD
- Stanford Sentiment Treebank (SST) (Correct answer)
- CoNLL-2003
- GLUE
Correct answer: Stanford Sentiment Treebank (SST)
SST provides phrase-level sentiment annotations on parse trees of movie reviews, supporting both binary and fine-grained 5-class evaluation.
Question 7: What is the purpose of 'sentiment-aware pre-training' in NLP models?
- Removing sentiment from embeddings to produce neutral representations
- Incorporating sentiment signals into the pretraining objective to produce embeddings more aligned with sentiment tasks (Correct answer)
- Reducing model size by pruning sentiment-related parameters
- Pretraining exclusively on positive text to bias the model toward positivity
Correct answer: Incorporating sentiment signals into the pretraining objective to produce embeddings more aligned with sentiment tasks
Sentiment-aware pretraining adds objectives (e.g., predicting review ratings) during pretraining so learned representations better capture affective content.
NLP Sentiment Analysis 5
What is the 'semantic orientation' of a word in sentiment analysis?