Topic Modeling Quiz

Answers and explanations appear as you take the quiz.

1. In the LDA formulation, how is a modeled topic represented over the vocabulary?
   • As one mandatory human-written keyword only
   • As a probability distribution over words
   • As a grammatical parse of every sentence
   • As a sentiment value with no vocabulary representation
2. Why can an unsupervised LDA topic model be fit to documents without topic labels?
   • It requires a verified category for every document first
   • It reads only documents with identical titles
   • It necessarily uses a classifier trained on the same task labels
   • It infers unobserved topic structure from the observed words
3. How does LDA treat an individual document?
   • As belonging to exactly one topic
   • As a sequence where word order is essential
   • As a mixture of multiple topics in different proportions
   • As an image of text to be scanned
4. What is the 'bag-of-words' assumption used by classic LDA?
   • It ignores word order and uses only word counts
   • It only keeps the first sentence of each document
   • It translates every document into English first
   • It requires words to be in alphabetical order
5. Which decision must you typically make before running LDA?
   • The exact title of each topic
   • The author of every document
   • The font the documents were written in
   • The number of topics K to find
6. Which stage commonly starts BERTopic's grouping of documents rather than relying only on raw word counts?
   • Assigning every document a manually required topic label
   • Clustering representations derived from document embeddings
   • Removing all text before computing any representation
   • Treating every document as an identical image file
7. When inspecting LDA's high-probability words, who can supply a descriptive topic name such as election coverage?
   • An analyst interpreting the weighted words and documents
   • A probability value that necessarily proves one official name
   • A required government approver for every model run
   • A document filename that always determines the meaning
8. After BERTopic groups documents, what does c-TF-IDF help produce for each cluster?
   • The original input image pixels
   • A required dependency parse for every token
   • Characteristic weighted words representing the group
   • Proof that every cluster has exactly one true theme