WorksheetsWeb Mining 4
Total questions: 112
Worksheet time: 56mins
Name
Class
Date
1.
What is the goal of information retrieval (IR) as described in the slides?
a)
Help users find information that matches their information needs
b)
Encrypt documents for secure storage
c)
Translate documents into multiple languages
d)
Compress video streams for faster playback
e)
Detect malware in executable files
2.
In the basic IR workflow, what does the user provide to the system?
a)
A user query
b)
A document index
c)
A vocabulary trie
d)
A postings list
e)
A compression codebook
3.
Which query type uses Boolean operators such as AND, OR, and NOT?
a)
Boolean query
b)
Keyword query
c)
Phrase query
d)
Proximity query
e)
Full document query
4.
According to the slides, which component indexes raw documents to enable efficient retrieval?
a)
Indexer
b)
User
c)
Query operations module
d)
Retrieval module
e)
Ranking interface
5.
In many IR models in the slides, documents and queries are represented as what?
a)
Bag of words
b)
Decision trees
c)
Ordered character sequences
d)
Graphs with labeled edges
e)
Audio waveforms
6.
In the Boolean model, how are documents retrieved after evaluating a Boolean query?
a)
All satisfying documents are returned without ranking
b)
Documents are ranked by cosine similarity
c)
Documents are ranked by PageRank only
d)
Documents are clustered into topics first
e)
Only the single best document is returned
7.
What does tf-idf aim to do compared with plain term frequency (tf)?
a)
Give lower weight to terms that appear in many documents
b)
Ignore term counts completely
c)
Use only hyperlink information
d)
Require exact phrase matching only
e)
Return documents without ranking
8.
Which similarity measure is stated as the most well known for ranking in the vector space model?
a)
Cosine similarity
b)
Edit distance
c)
Hamming distance
d)
Earth mover distance
e)
Jaccard similarity
9.
In the statistical language model approach, documents are ranked by what?
a)
The likelihood of the query given the document language model
b)
The number of hyperlinks pointing to the document
c)
The alphabetical order of document titles
d)
The total file size of the document
e)
The order documents were added to the collection
10.
In the unigram language model described, what is the purpose of smoothing?
a)
Prevent zero probabilities for unseen terms in a document
b)
Force all terms to have equal probability
c)
Remove stopwords from the query
d)
Compute PageRank scores
e)
Convert documents to lowercase
11.
In relevance feedback, what does the user do after the initial retrieval?
a)
Identify some relevant and irrelevant documents
b)
Rewrite the entire document collection
c)
Delete the document index
d)
Disable query expansion permanently
e)
Sort documents alphabetically
12.
In the Rocchio method, the expanded query is formed using which sets of documents?
a)
Relevant and irrelevant documents selected by the user
b)
Only the longest documents in the collection
c)
Only documents that contain stopwords
d)
Only documents with the highest PageRank
e)
Only documents with duplicated content
13.
In the machine learning method for relevance feedback, test documents are assigned to which class?
a)
The class whose vector has the highest cosine similarity
b)
The class with the fewest training examples
c)
The class with the smallest document length
d)
The class with the most stopwords
e)
The class with the lowest term frequency
14.
PU learning in the slides refers to learning from which types of examples?
a)
Positive and unlabeled examples
b)
Positive and negative examples only
c)
Unlabeled examples only
d)
Perfectly balanced labeled examples
e)
Randomly generated synthetic examples only
15.
Pseudo-relevance feedback assumes what about the top-ranked documents?
a)
They are relevant
b)
They are spam
c)
They are duplicates
d)
They are all irrelevant
e)
They are always the newest documents
16.
Precision at rank position i is defined using which quantity in the slides?
a)
The number of relevant documents found in the top i results divided by i
b)
The number of retrieved documents divided by the collection size
c)
The number of query terms divided by i
d)
The number of documents in the vocabulary divided by i
e)
The number of stopwords divided by i
17.
What does average precision summarize for a ranked list?
a)
Precision values at ranks where relevant documents occur
b)
Only the recall value at the final rank
c)
Only the number of non-relevant documents
d)
Only the highest precision value observed
e)
Only the document length distribution
18.
A precision-recall curve relates precision to which other measure?
a)
Recall
b)
Document length
c)
Index size
d)
Vocabulary size
e)
Crawl depth
19.
What does the F-score combine according to the slides?
a)
Precision and recall
b)
Precision and PageRank
c)
Recall and document length
d)
Index size and query time
e)
Term frequency and stopword count
20.
Which of the following is listed as a text preprocessing step?
a)
Stemming
b)
Hyperlink analysis
c)
PageRank computation
d)
Spam farm creation
e)
Authority score boosting
21.
Stopwords are described as words that are typically what?
a)
Frequently occurring and insignificant for document content
b)
Rare and highly discriminative for topics
c)
Always numerical tokens
d)
Always proper nouns
e)
Only found in URLs
22.
What is stemming in the context of text preprocessing?
a)
Reducing words to their stems or roots
b)
Removing all punctuation from documents only
c)
Replacing every word with its synonym
d)
Sorting tokens alphabetically
e)
Keeping only uppercase words
23.
Which preprocessing operation converts letters into a single form?
a)
Case normalization
b)
Link analysis
c)
Anchor text extraction
d)
Query likelihood estimation
e)
Spam cloaking detection
24.
In web page preprocessing, why are terms in a page title often given higher weight?
a)
The title is usually a concise description of the page
b)
The title always contains more words than the body
c)
Titles are generated randomly by crawlers
d)
The title field is invisible to users
e)
Titles are always duplicated across pages
25.
A common web page preprocessing goal is to identify which part of a page?
a)
The main content blocks
b)
Only the background colors
c)
Only the exact pixel coordinates of images
d)
Only the browser window size
e)
Only the font family used
26.
Which technique can detect exact duplicate pages according to the slides?
a)
Hashing or checksum
b)
Stemming
c)
Query expansion
d)
Cosine similarity on rankings
e)
Borda count
27.
What is the baseline approach to answering a query that is said to be inefficient?
a)
Scanning every document in the database for the query terms
b)
Using an inverted index
c)
Using a trie to store vocabulary
d)
Using gap encoding for document IDs
e)
Using PageRank to prefilter documents
28.
An inverted index associates each term with what?
a)
A list of documents that contain the term
b)
A list of stopwords to remove
c)
A list of HTML tags
d)
A list of search engines to query
e)
A list of compression algorithms only
29.
In a postings list entry, which piece of information can record where a term occurs within a document?
a)
Offsets (positions) of the term
b)
The document title font size
c)
The number of outbound links
d)
The page background color
e)
The crawler user-agent string
30.
When searching an inverted index, which step finds each query term in the vocabulary structure?
a)
Vocabulary search
b)
Results merging
c)
Rank score computation
d)
Spam filtering
e)
Link exchange detection
31.
In the example query processing, why can a document get a higher relevance score for a two-term query?
a)
Because the query terms appear near each other and in the same order
b)
Because the document is the oldest in the collection
c)
Because the document has the most stopwords
d)
Because the document has the fewest words
e)
Because the document has the largest file size
32.
Which data structure is mentioned for index construction with complexity proportional to vocabulary size?
a)
Trie
b)
Stack
c)
Heap
d)
Queue
e)
Bloom filter
33.
When pages are constantly added or deleted, the slides suggest maintaining additional indexes for what?
a)
Added pages and deleted pages
b)
Only stopwords and punctuation
c)
Only titles and metadata
d)
Only anchor text and URLs
e)
Only duplicate pages
34.
Why is gap encoding useful for compressing postings lists?
a)
Gaps between sorted document IDs are smaller and need fewer bits
b)
It makes document IDs random
c)
It increases vocabulary size
d)
It forces all postings lists to have equal length
e)
It removes the need for indexing
35.
Which coding method is described as efficient for small integers?
a)
Elias gamma coding
b)
Elias delta coding
c)
PageRank coding
d)
Stemming coding
e)
Anchor-text coding
36.
Golomb coding represents an integer using what two parts?
a)
A unary-coded quotient and a binary-coded remainder
b)
Two independent cosine similarities
c)
A stopword list and a stem list
d)
A PageRank value and a hub score
e)
A URL and a title string
37.
In Golomb decoding, what is computed first from the unary part?
a)
The quotient value
b)
The vocabulary size
c)
The document length
d)
The cosine similarity
e)
The list of offsets
38.
In variable-byte coding, what is special about the last bit of each byte as described?
a)
It indicates whether this byte is the last one for the integer
b)
It stores the entire integer value
c)
It always equals one
d)
It is used to encode stopwords
e)
It stores a PageRank score
39.
Latent Semantic Indexing (LSI) is introduced to address which problem?
a)
Relevant documents may use synonyms different from the query words
b)
Documents are always too short
c)
Queries cannot contain stopwords
d)
Hyperlinks cannot be crawled
e)
All documents are exact duplicates
40.
SVD factorizes a matrix into the product of how many matrices?
a)
Three
b)
Two
c)
Four
d)
Five
e)
One
41.
Using the k-largest singular components in SVD is intended to do what?
a)
Approximate the original matrix while removing noise
b)
Increase the rank of the matrix
c)
Make every term weight identical
d)
Force exact phrase matching
e)
Eliminate the need for any indexing
42.
In the SVD interpretation slide, directions with small variation among documents are treated as what?
a)
Noise that can be removed
b)
The most important topic signals
c)
Mandatory stopword dimensions
d)
Exact duplicates
e)
Authority links
43.
In LSI retrieval, where is the query transformed before comparison?
a)
Into the k-concept space
b)
Into a stopword list
c)
Into a URL-only representation
d)
Into a boolean expression only
e)
Into a checksum table
44.
In the LSI example, the term-document data are organized into what type of matrix?
a)
A term-document matrix
b)
A confusion matrix
c)
A covariance matrix of images
d)
A graph adjacency matrix of hyperlinks
e)
A bytecode instruction matrix
45.
In SVD, which matrix contains the singular values arranged on a diagonal?
a)
Sigma
b)
U
c)
V
d)
A
e)
D
46.
In SVD, what do the columns of V represent as described in the slides?
a)
Right singular vectors
b)
Stopword lists
c)
Document IDs
d)
Anchor text strings
e)
Hash signatures
47.
After transforming into concept space, LSI compares the query to documents using which type of function?
a)
A similarity function such as cosine similarity
b)
Only exact string matching
c)
Only URL comparison
d)
Only checksum equality
e)
Only document length sorting
48.
Which limitation of LSI is mentioned as making it unsuitable for large-scale web search?
a)
High computational complexity
b)
It cannot handle any synonyms
c)
It requires manual crawling of every page
d)
It can only work with Boolean queries
e)
It always returns unranked results
49.
Which sequence best matches the web search process listed in the slides?
a)
Crawl, parse, index, store, query, retrieve
b)
Query, retrieve, crawl, parse, store, index
c)
Parse, crawl, retrieve, store, index, query
d)
Store, index, crawl, query, parse, retrieve
e)
Retrieve, store, parse, query, crawl, index
50.
In web search ranking, a link from page x to page y is treated as what?
a)
An implicit conveyance of authority from x to y
b)
A guarantee that y is spam
c)
A duplicate marker
d)
A stopword signal
e)
A document deletion request
51.
What is the main purpose of a meta-search engine as described?
a)
Combine results of multiple search engines
b)
Replace indexing with hashing only
c)
Eliminate ranking completely
d)
Perform only stemming and stopword removal
e)
Detect exact duplicates using checksums only
52.
CombMAX in meta-search combination returns what for a document?
a)
The maximum similarity score across engines
b)
The minimum similarity score across engines
c)
The average rank position across engines
d)
The number of engines that did not return it
e)
A binary value indicating exact duplication
53.
In reciprocal ranking, what score does the top document receive from an engine?
a)
1 point
b)
0 points
c)
2 points
d)
Half a point
e)
A negative point
54.
Which combination method is based on pair-wise comparisons of documents across engines?
a)
Condorcet ranking
b)
CombSUM
c)
CombMAX
d)
Hashing
e)
Stemming
55.
According to the slides, why do spammers try to achieve high rankings in search results?
a)
High-ranking websites can bring popularity and financial benefits
b)
Search engines require every site to spam to be indexed
c)
High rank reduces the need for content
d)
High rank prevents crawling
e)
High rank removes duplicates automatically
56.
Which content spamming technique increases a document's term frequency scores by repeating terms?
a)
Repeating important terms
b)
Removing stopwords
c)
Applying stemming rules
d)
Using gap encoding
e)
Computing average precision
57.
What is a honey pot in link spamming as described?
a)
Useful-looking pages that attract links but contain hidden links to target spam pages
b)
A checksum used to detect duplicates
c)
A compressed postings list
d)
A query expansion rule
e)
A page title with fewer keywords
58.
Cloaking is described as doing what?
a)
Returning different content to users and to web crawlers
b)
Removing HTML tags from a page
c)
Indexing only the title of a page
d)
Deleting all out-links from a page
e)
Replacing hyphens with spaces
59.
Which feature is mentioned for detecting content spam pages?
a)
Fraction of visible content
b)
RGB color palette of the logo
c)
Exact pixel position of the banner
d)
Screen resolution of the viewer
e)
Browser tab order
60.
In the basic IR architecture, why is an index built before retrieval?
a)
To enable efficient retrieval without scanning every document
b)
To guarantee that every retrieved document is relevant
c)
To remove all synonyms from the collection
d)
To force all queries to be Boolean queries
e)
To compute PageRank scores for every web page
61.
Compared with a keyword query, what additional constraint does a phrase query impose?
a)
Returned documents must contain the words as a contiguous phrase
b)
Returned documents must contain the words in any order anywhere
c)
Returned documents must be ranked only by hyperlink distance
d)
Returned documents must be exact duplicates of the query
e)
Returned documents must contain no stopwords
62.
If an IR model represents text as a bag of words, which information is typically ignored?
a)
The order and positions of terms
b)
Whether the document is indexed
c)
Whether the query contains terms
d)
The size of the document collection
e)
The fact that documents are stored
63.
Why can the Boolean model be inconvenient for users in many search scenarios?
a)
It returns matching documents without a ranking order
b)
It cannot express OR conditions
c)
It requires document length normalization
d)
It needs smoothing to avoid zero probabilities
e)
It cannot use an inverted index
64.
Why does tf-idf reduce the weight of terms that appear in many documents?
a)
Such terms are less discriminative for distinguishing relevant documents
b)
Such terms always indicate spam content
c)
Such terms cannot be stored in the vocabulary
d)
Such terms are removed as punctuation
e)
Such terms only appear in titles
65.
The slides note that Okapi is more efficient for which scenario?
a)
Short query retrieval
b)
Full-document queries only
c)
Queries containing only stopwords
d)
Queries where all terms are equally frequent
e)
Queries that require exact duplicates
66.
In a query-likelihood language model, what problem arises if a query term never appears in a document and no smoothing is used?
a)
The document can receive zero likelihood for the whole query
b)
The document automatically becomes the top-ranked result
c)
The document is treated as a duplicate
d)
The query terms are automatically removed as stopwords
e)
The document length becomes negative
67.
How can relevance feedback improve retrieval results over a second round?
a)
By expanding or redefining the query using terms from judged documents
b)
By deleting the document index and rebuilding it randomly
c)
By forcing all documents to be returned without ranking
d)
By replacing the query with a checksum
e)
By removing all HTML tags from the query
68.
In Rocchio-style feedback, what is the intended effect of incorporating user-labeled irrelevant documents?
a)
Reduce the influence of terms associated with irrelevant documents
b)
Increase the weight of every term in the original query
c)
Make the query purely Boolean
d)
Guarantee perfect precision at rank 1
e)
Eliminate the need for any document representation
69.
In the relevance-feedback classification approach described, why is cosine similarity used between a test document and class vectors?
a)
It measures similarity between vector representations of documents and classes
b)
It counts the number of stopwords shared
c)
It identifies exact duplicate pages only
d)
It computes hyperlink authority scores
e)
It normalizes URLs into one case
70.
What distinguishes LU learning from PU learning in the slides?
a)
LU uses both relevant and irrelevant labels plus unlabeled examples, while PU uses only positive labels plus unlabeled examples
b)
LU uses only positive labels, while PU uses only negative labels
c)
LU requires full-document queries, while PU forbids them
d)
LU is for Boolean queries, while PU is for phrase queries
e)
LU removes stopwords, while PU removes punctuation
71.
Pseudo-relevance feedback can fail when its key assumption is violated. What assumption is that?
a)
Top-ranked documents are relevant
b)
Stopwords are always present in queries
c)
Documents are always represented as graphs
d)
All documents are duplicates
e)
All hyperlinks are spam
72.
The slides state that in practice high recall is often achieved at the expense of what?
a)
Precision
b)
Query length
c)
Vocabulary size
d)
Document length
e)
Index compression ratio
73.
Why is it often difficult to determine the set of relevant documents for a query on the Web?
a)
The set of relevant documents can be extremely large and hard to enumerate
b)
Web pages cannot be indexed
c)
Queries cannot contain more than one term
d)
Relevance feedback is mandatory for all searches
e)
Search engines always return unranked results
74.
If stopwords are removed from both documents and queries, which result is most directly expected?
a)
The index focuses on more content-bearing terms
b)
Phrase queries become impossible in principle
c)
All documents receive identical relevance scores
d)
Vocabulary size becomes exactly zero
e)
All duplicate pages are automatically detected
75.
Stemming can increase recall but also return irrelevant documents. Which explanation matches the slides?
a)
Different words can be reduced to the same stem and be conflated
b)
Stemming removes all digits from all documents
c)
Stemming forces exact phrase matching
d)
Stemming replaces every word with a synonym
e)
Stemming guarantees that tf-idf weights are zero
76.
According to the slides, how do traditional IR systems and search engines often differ in handling digits?
a)
Traditional IR often removes digits, while search engines usually index them
b)
Traditional IR always indexes digits, while search engines always remove them
c)
Both always remove digits without exception
d)
Both always treat digits as stopwords only
e)
Both require digits to be converted into hyphens
77.
Why is anchor text treated specially in search engines?
a)
It often provides an accurate description of the linked page's content
b)
It is always longer than the body text
c)
It is invisible to crawlers
d)
It cannot be parsed from HTML
e)
It is used only for checksum computation
78.
Why do web search systems try to identify main content blocks in a web page?
a)
To reduce the influence of ads, navigation bars, and other non-main content
b)
To increase the number of stopwords indexed
c)
To remove all hyperlinks from the page
d)
To ensure every page has the same title length
e)
To force all pages into a single template
79.
If two pages differ by small edits but share substantial text, which approach from the slides is more suitable than hashing?
a)
Using n-grams (shingles) with Jaccard similarity
b)
Using a checksum only
c)
Using Boolean retrieval only
d)
Using case normalization only
e)
Using reciprocal ranking only
80.
Why does an inverted index speed up query processing compared with scanning all documents?
a)
It lets the system access only documents that contain query terms
b)
It forces every term to appear in every document
c)
It removes the need for any vocabulary
d)
It guarantees perfect recall for every query
e)
It ranks documents without any scoring function
81.
A postings list entry can include frequency and offsets of a term. Which type of query benefits most from having offsets?
a)
Phrase or proximity queries
b)
Boolean queries only
c)
Full-document queries only
d)
Queries that ignore term occurrence
e)
Queries that return all documents unranked
82.
The slides mention storing the vocabulary in structures such as hash tables, tries, or B-trees. What is the purpose of these structures?
a)
To speed up finding query terms in the vocabulary
b)
To compute average precision
c)
To remove duplicate URLs
d)
To identify spam farms
e)
To smooth language models
83.
During results merging in an inverted index, why does the process often start with the shortest postings list?
a)
It reduces the number of candidate documents that must be checked in other lists
b)
It increases the vocabulary size
c)
It guarantees that the first document is relevant
d)
It avoids the need for caching
e)
It forces all documents to be returned
84.
When maintaining separate indexes for added and deleted pages, how should final results conceptually be formed?
a)
Include documents from the main and added indexes, excluding deleted ones
b)
Include only deleted documents
c)
Exclude both main and added indexes
d)
Rank documents only by checksum
e)
Return only pages with many stopwords
85.
Why does storing gaps between adjacent document IDs improve compression?
a)
Gaps are typically smaller numbers than the original IDs
b)
Gaps are always larger and easier to encode
c)
Gaps remove the need to store term frequencies
d)
Gaps make postings lists unsorted
e)
Gaps eliminate the need for decoding
86.
The slides note that variable-byte coding is often faster than variable-bit schemes despite higher storage cost. What is the key trade-off?
a)
Faster decoding versus less compact storage
b)
More compact storage versus slower crawling
c)
Better stemming versus worse stopword removal
d)
Higher PageRank versus lower authority
e)
More duplicates versus fewer duplicates
87.
In LSI, why is the query transformed into concept space before retrieval?
a)
Because retrieval compares concepts rather than original terms
b)
Because concept space is identical to the stopword list
c)
Because concept space stores URLs only
d)
Because concept space eliminates the need for ranking
e)
Because concept space forces Boolean matching only
88.
What is the intended effect of choosing a smaller k in truncated SVD for LSI?
a)
Remove more noise but risk losing some information
b)
Increase noise while keeping all information
c)
Guarantee that concepts are perfectly interpretable
d)
Make computation faster while increasing matrix rank
e)
Ensure exact duplication detection
89.
Why can hyperlink information help improve web search results beyond content relevance alone?
a)
Links can act as implicit endorsements of page quality
b)
Links guarantee that pages contain the query terms
c)
Links remove the need for indexing
d)
Links force all pages to be spam
e)
Links make document length irrelevant
90.
What is one reason meta-search can improve effectiveness according to the slides?
a)
It combines multiple rankings to increase coverage and quality
b)
It prevents crawlers from visiting pages
c)
It converts all queries into proximity queries
d)
It eliminates duplicate detection entirely
e)
It removes the need for relevance scoring
91.
Compared with CombSUM, what additional factor does CombMNZ use that can boost a document's combined score?
a)
The number of search engines that returned a non-zero score for the document
b)
The number of stopwords in the document
c)
The number of HTML tags in the page
d)
The total number of documents in the web
e)
The file size of the query
92.
A retrieval system uses a pure bag-of-words representation and does not store term positions. Which capability is most directly limited by this design?
a)
Reliable phrase and proximity matching
b)
Stopword removal
c)
Vocabulary construction
d)
Document ID assignment
e)
Basic keyword matching for single terms
93.
In a query-likelihood language model, why can smoothing change the relative ranking between two documents when a query contains a rare term?
a)
It prevents any document from getting zero probability due to missing terms, so other evidence can still affect ranking
b)
It forces all rare terms to be removed as stopwords
c)
It replaces cosine similarity with PageRank
d)
It guarantees that the shortest document always wins
e)
It converts the model into the Boolean model
94.
Suppose a term appears frequently in both user-marked relevant and user-marked irrelevant documents. In Rocchio-style feedback, what is the most likely net effect on that term's weight in the expanded query?
a)
It may be reduced because the irrelevant-document contribution subtracts from the relevant-document contribution
b)
It must always increase because it appears in relevant documents
c)
It becomes a stopword automatically
d)
It is replaced by its stem regardless of feedback
e)
It is ignored because Rocchio uses only Boolean logic
95.
In web search, users may click only a few results based on snippets. According to the slides' PU-learning framing, what is the most accurate interpretation of unclicked results?
a)
They are unlabeled rather than confirmed irrelevant
b)
They are confirmed irrelevant examples
c)
They are confirmed positive examples
d)
They are duplicates that must be removed
e)
They are pages that cannot be indexed
96.
Pseudo-relevance feedback expands a query using terms from top-ranked documents. Which situation most directly risks query drift based on the slide assumptions?
a)
The top-ranked documents are not actually relevant to the user's need
b)
The query is short
c)
The collection uses an inverted index
d)
Stopwords are removed from documents
e)
Documents are represented as vectors
97.
If it is nearly impossible to enumerate all relevant documents for a web query, which evaluation approach from the slides is most practical to report?
a)
Rank precision at fixed cutoffs such as P(10)
b)
Exact recall over the entire web
c)
Average document length of relevant documents
d)
The full set of relevant documents as ground truth
e)
The exact precision-recall curve for all recall levels without judgments
98.
A site copies another page but inserts and deletes small pieces of text throughout. Based on the duplicate-detection slide, which method is the best fit?
a)
Shingles (n-grams) combined with Jaccard similarity
b)
Hashing the whole page as a single string
c)
Unary coding of document IDs
d)
Boolean retrieval with NOT only
e)
Reciprocal rank fusion
99.
When postings lists are stored mostly on disk, which merging strategy from the slides helps minimize work for an AND query across multiple terms?
a)
Iterate the shortest postings list and probe others for matches
b)
Concatenate all postings lists and sort them each time
c)
Scan every document and check the terms directly
d)
Compute PageRank for all pages before merging
e)
Ignore the vocabulary and search raw text only
100.
The slides mention hash tables, tries, and B-trees for vocabulary search. If you want space efficiency via lexicographic order and still allow fast lookup, what additional method is suggested?
a)
Binary search over the lexicographically ordered vocabulary
b)
Compute tf-idf for every term before lookup
c)
Use Laplace smoothing on vocabulary entries
d)
Apply stemming to the entire vocabulary on every query
e)
Use reciprocal ranking to order vocabulary terms
101.
An index is maintained with a main collection, an added-pages index, and a deleted-pages index. Which operation best matches the final result set described in the slides?
a)
Union the main and added sets, then remove the deleted set
b)
Intersect the main and deleted sets, then add the added set
c)
Use only the deleted set because it is most recent
d)
Subtract the added set from the main set
e)
Take the symmetric difference of all three sets
102.
Gap encoding relies on document IDs being stored in increasing order. If postings were not sorted, what would most likely happen to compression effectiveness?
a)
It would generally worsen because gaps would not stay small and predictable
b)
It would improve because gaps would become larger
c)
It would be unchanged because order never matters
d)
It would eliminate the need for any decoding
e)
It would turn variable-byte coding into unary coding
103.
Based on the slides, which pairing best matches coding choice to the typical integer size it targets?
a)
Gamma for small integers, delta for larger integers
b)
Gamma for large integers, delta for small integers
c)
Both are only for floating-point values
d)
Both require document IDs to be unsorted
e)
Delta is only used for stopword lists
104.
If your primary goal is faster decoding during retrieval even at the cost of extra storage, which compression family from the slides is most aligned with that goal?
a)
Variable-byte coding
b)
Elias gamma coding only
c)
Elias delta coding only
d)
Unary coding only
e)
Checksum hashing of postings
105.
Golomb coding uses a parameter b. The slides relate b to how many documents contain a term. If a term appears in fewer documents, what does the rule imply about b?
a)
b should be larger
b)
b should be smaller
c)
b should be exactly one
d)
b is independent of term frequency across documents
e)
b must equal the vocabulary size
106.
Which combination of factors from the LSI discussion most strongly explains why LSI is not suitable for web search at scale?
a)
High computational cost and difficulty choosing k by trial and error
b)
Inability to represent documents as bags of words
c)
Requirement that all queries be Boolean
d)
Need for exact duplicate detection before indexing
e)
Dependence on hyperlink authority only
107.
In truncated SVD for LSI, what is a realistic risk of setting k too small compared with the original matrix?
a)
Important term associations may be lost along with noise
b)
The concept space becomes perfectly interpretable
c)
Computation becomes slower because rank increases
d)
The method becomes identical to hashing
e)
The vocabulary size necessarily increases
108.
The SVD interpretation slide suggests removing directions with small variance. What is the underlying assumption behind removing those directions?
a)
Low-variance directions mainly reflect noise rather than meaningful semantic structure
b)
Low-variance directions always contain the query terms
c)
High-variance directions are always spam signals
d)
All directions are equally informative in retrieval
e)
Variance depends only on document length, not content
109.
A ranking function uses term proximity and term order as content factors. Which scenario should receive the highest boost under that design, consistent with the slides?
a)
Documents where query terms appear close together and in the same sequence as the query
b)
Documents where query terms appear far apart in reverse order
c)
Documents where only stopwords match the query
d)
Documents that contain the query terms only in HTML tags
e)
Documents that contain no query terms but have many links
110.
In similarity-score combination for meta-search, CombMIN takes the minimum score across engines. Which behavior does this most directly encourage?
a)
A document is penalized if any engine scores it very low
b)
A document is rewarded even if only one engine returns it
c)
A document's score depends only on rank positions
d)
A document is boosted for being returned by many engines regardless of score
e)
A document is scored by pair-wise voting only
111.
In Condorcet ranking as described, when a search engine does not return a document, how does that affect pair-wise comparisons involving that document within that engine's results?
a)
Ranked documents are treated as beating unranked documents
b)
Unranked documents are treated as beating ranked documents
c)
All unranked documents are treated as tied above ranked ones
d)
Only the top-ranked document participates in comparisons
e)
Pair-wise comparisons are not used in Condorcet ranking
112.
Out-link spamming tries to add links to authoritative pages to boost hub scores. If a search engine also uses link-based authority signals, what is the most plausible limitation of relying only on link structure for quality?
a)
Link patterns can be manipulated to inflate reputation signals without adding real informational value
b)
Link structure automatically detects duplicate content perfectly
c)
Link structure guarantees high precision and recall for all queries
d)
Link structure makes stopword removal unnecessary
e)
Link structure prevents cloaking by design
100 %
