NLP Concepts DY0-001 Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A data science team is building a recommendation system for academic articles. They have two options: (1) use pre-trained word embeddings from a general corpus, or (2) train custom embeddings on their 2-million-article corpus. The corpus contains highly specialized terminology not well-represented in general text. The team has limited compute resources and must balance accuracy against training time. What should they do?

   • AUse TF-IDF vectors instead of embeddings
   • BTrain embeddings on a random 10% sample of the corpus
   • CUse pre-trained general embeddings to save compute time
   • DTrain custom embeddings on the specialized corpus, accepting longer training time
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A news aggregator needs to identify the main subject of each article. The team notices that the word 'the' appears in every article and 'election' appears in only 5% of articles. They want to reduce noise before analysis. Which preprocessing step combined with which weighting scheme would best highlight the subject?

   • AApply TF-IDF without removing stop words
   • BRemove stop words, then apply TF-IDF
   • CUse word embeddings without preprocessing
   • DTokenize with byte-pair encoding only
   Reveal Solution Discussion Select an answer first
4. 18expert · hard

   A social media analytics company processes tweets that contain emojis, hashtags, and URLs. They want to build a sentiment model, but the raw text is noisy. They must decide between (1) removing all non-alphanumeric characters, or (2) replacing emojis with sentiment tokens (e.g., ':)' → 'POSITIVE_EMOJI') while keeping hashtags. Which approach better preserves sentiment information?

   • ARemove all non-alphanumeric characters
   • BReplace emojis with sentiment tokens and keep hashtags
   • CReplace hashtags with the word 'hashtag' and remove emojis
   • DRemove emojis but keep hashtags as single tokens
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is the key property of word embeddings in a continuous vector space?

   • AThe vector dimensions correspond to predefined grammatical categories
   • BWords with similar meanings are located near each other in the vector space
   • CEach word is represented by a unique one-hot encoded vector
   • DAll words are mapped to the same vector regardless of context
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What does the inverse document frequency (IDF) component of TF-IDF measure?

   • AHow rare or common a word is across the entire corpus
   • BHow often a word appears in a single document
   • CThe length of each document in the corpus
   • DThe total number of documents in the corpus
   Reveal Solution Discussion Select an answer first