Font size
S
M
L
XL
WorksheetsWeb Mining 4: Information Retrieval
Total questions: 109
Worksheet time: 55mins
Name
Class
Date
1.
Which component of an information retrieval system is primarily responsible for calculating a relevance score for documents against a query?
a)
Query Processor
b)
Indexer
c)
Retrieval Component
d)
Document Collection
e)
User Interface
2.
In a typical information retrieval system diagram, what is the primary input to the 'Query Processor' from the user?
a)
A set of ranked documents
b)
A relevance feedback signal
c)
A query
d)
A document collection
e)
An inverted index
3.
What is the main purpose of creating a text index from a document collection in an IR system?
a)
To compress the document collection for storage
b)
To improve the efficiency of query processing
c)
To perform semantic analysis on the documents
d)
To remove stop words and perform stemming
e)
To directly calculate document relevance scores
4.
Which query type explicitly uses operators like AND, OR, and NOT to combine keywords?
a)
Keyword Query
b)
Phrase Query
c)
Proximity Query
d)
Boolean Query
e)
Natural Language Query
5.
A query like "information retrieval systems" where the exact sequence of words must be present in the returned documents is best described as what type of query?
a)
Keyword Query
b)
Phrase Query
c)
Boolean Query
d)
Proximity Query
e)
Document Query
6.
What distinguishes a Proximity Query from a standard Keyword Query?
a)
It requires all keywords to be present in the document.
b)
It ranks documents based on the closeness of query keywords within them.
c)
It uses logical operators like AND and OR.
d)
It accepts a full document as the query input.
e)
It only returns documents containing an exact phrase match.
7.
During query processing, what is the primary goal of removing 'stop words'?
a)
To reduce the query to its grammatical root form.
b)
To improve recall by expanding the query with synonyms.
c)
To increase processing speed by eliminating common, low-meaning words.
d)
To convert the query into a machine-executable format.
e)
To ensure the word order of the original query is preserved.
8.
According to the slides, which data structure is described as highly effective and easy to apply for indexing text documents?
a)
B-Tree
b)
Hash Table
c)
Trie
d)
Inverted Index
e)
Suffix Array
9.
In the retrieval component's process, when are relevance scores for documents typically calculated?
a)
Before the query is processed.
b)
Only for documents containing at least one of the query keywords.
c)
For all documents in the entire collection.
d)
After the user provides relevance feedback.
e)
During the text indexing phase.
10.
In Information Retrieval models, what does the 'bag of words' representation assumption ignore?
a)
The frequency of words
b)
The length of the document
c)
The order and position of words
d)
The presence of stop words
e)
The total number of unique words in the collection
11.
In the formal notation of IR models, if V is the vocabulary of a document collection D, what does |V| represent?
a)
The total number of documents in the collection.
b)
The average number of words per document.
c)
The number of unique words (vocabulary size).
d)
The number of times a specific term appears.
e)
The total number of words in the collection.
12.
In the Boolean Model of information retrieval, how are the results presented to the user?
a)
As a ranked list based on relevance scores.
b)
As an unranked set of documents that satisfy the query logic.
c)
As a list sorted by document length.
d)
As a list ranked by the number of query terms matched.
e)
As a probability distribution over the document collection.
13.
What is the primary limitation of the traditional Boolean model for information retrieval?
a)
It cannot handle queries with logical operators like AND or NOT.
b)
It is computationally very expensive for large collections.
c)
It does not provide a mechanism for ranking retrieved documents by relevance.
d)
It requires documents to be represented as dense vectors.
e)
It cannot be used with an inverted index structure.
14.
In the vector space model, what does the term frequency (tf) of a term t in a document d represent?
a)
A normalized value of the term's importance in the collection.
b)
The number of times term t appears in document d.
c)
A logarithmically scaled count of documents containing term t.
d)
A binary value indicating the presence or absence of term t.
e)
The inverse of the number of documents containing term t.
15.
What is the primary purpose of the Inverse Document Frequency (idf) component in the tf-idf weighting scheme?
a)
To increase the weight of terms that appear frequently in a single document.
b)
To decrease the weight of terms that are common across many documents.
c)
To normalize the term frequency by the maximum frequency in the document.
d)
To ensure all term weights are between 0 and 1.
e)
To increase the weight of terms that are rare across the entire collection.
16.
The formula for idf is log(N / df_i). What does df_i represent in this context?
a)
The frequency of term i in a specific document.
b)
The total number of documents in the collection.
c)
The number of documents in the collection that contain term i.
d)
The length of document i.
e)
The maximum term frequency in the collection.
17.
Which similarity measure is most commonly used in the Vector Space Model to calculate the relevance between a query vector and a document vector?
a)
Euclidean Distance
b)
Jaccard Similarity
c)
Manhattan Distance
d)
Okapi BM25
e)
Cosine Similarity
18.
The Okapi BM25 ranking function is noted to be more effective than standard cosine similarity for what type of queries?
a)
Long, descriptive queries
b)
Boolean queries
c)
Short queries
d)
Phrase queries
e)
Queries with many stop words
19.
In the Statistical Language Model for IR, what fundamental probability is estimated to rank documents?
a)
The probability of a document being relevant, Pr(relevant|d).
b)
The probability of a document given a query, Pr(d|q).
c)
The prior probability of a query, Pr(q).
d)
The probability that a document's language model would generate the query, Pr(q|d).
e)
The joint probability of the query and document, Pr(q, d).
20.
What is the purpose of 'smoothing' in a statistical language model for IR?
a)
To increase the probability of common words.
b)
To ensure the sum of probabilities is exactly 1.
c)
To avoid zero probabilities for query words not present in a document.
d)
To normalize document lengths.
e)
To convert term frequencies into probabilities.
21.
What user action initiates the Relevance Feedback process?
a)
Submitting an initial query.
b)
Selecting which documents from the initial results are relevant or not.
c)
Revising the query with new keywords manually.
d)
Clicking on the second page of search results.
e)
Choosing a different IR model.
22.
In the Rocchio algorithm for query expansion, what is the role of the documents identified as non-relevant by the user?
a)
They are added to the query vector to improve recall.
b)
They are ignored, as only relevant documents are used for expansion.
c)
They are used to move the query vector away from the non-relevant document space.
d)
They are used to calculate the initial query vector's magnitude.
e)
They are used to determine the optimal number of expansion terms.
23.
In the Rocchio algorithm formula, what does the 'alpha' parameter typically control?
a)
The weight of the non-relevant documents.
b)
The weight of the relevant documents.
c)
The weight of the original query vector.
d)
The number of expansion terms to add.
e)
The normalization factor for document vectors.
24.
What is a key difference between standard Relevance Feedback and Pseudo-Relevance Feedback?
a)
Pseudo-relevance feedback uses machine learning classifiers.
b)
Pseudo-relevance feedback does not modify the original query.
c)
Pseudo-relevance feedback requires explicit user judgments.
d)
Pseudo-relevance feedback assumes the top-ranked documents are relevant without user input.
e)
Pseudo-relevance feedback can only be used with the Boolean model.
25.
What is the main assumption made in Pseudo-Relevance Feedback?
a)
That all documents in the collection are equally relevant.
b)
That users always click on the first result.
c)
That the initial top-k retrieved documents are relevant.
d)
That query terms should be expanded using a thesaurus.
e)
That non-relevant documents are more valuable for feedback than relevant ones.
26.
In IR evaluation, what does Precision at position i, p(i), measure?
a)
The total number of relevant documents found up to position i.
b)
The proportion of retrieved documents up to position i that are relevant.
c)
The proportion of all existing relevant documents that have been found up to position i.
d)
The average relevance score of the top i documents.
e)
The probability that the document at position i is relevant.
27.
In IR evaluation, what does Recall at position i, r(i), measure?
a)
The proportion of retrieved documents up to position i that are relevant.
b)
The total number of documents retrieved up to position i.
c)
The proportion of all existing relevant documents that have been retrieved up to position i.
d)
The inverse of precision at position i.
e)
The number of relevant documents in the top i results.
28.
Average Precision (AP) for a single query is calculated by averaging the precision values at what specific points?
a)
At every retrieved document position.
b)
At every 10% increment of recall.
c)
Only at the positions of the non-relevant documents.
d)
Only at the positions where a relevant document is retrieved.
e)
At a fixed set of positions, such as 5, 10, and 15.
29.
What is the typical relationship observed between precision and recall in an information retrieval system?
a)
They are directly proportional; as one increases, the other also increases.
b)
They are independent of each other.
c)
There is often a trade-off; increasing recall tends to decrease precision.
d)
Precision is always higher than recall.
e)
Recall is always higher than precision.
30.
The F-score (or F-measure) is a metric that combines which two other evaluation measures?
a)
Average Precision and Mean Average Precision
b)
Precision and Recall
c)
Accuracy and Error Rate
d)
Latency and Throughput
e)
Precision at k and Recall at k
31.
Why are metrics like P@10 (Precision at 10) particularly useful for evaluating web search engines?
a)
Because web search engines can determine the total number of relevant documents.
b)
Because most users rarely look beyond the first page of results.
c)
Because it is a better measure of overall recall than Average Precision.
d)
Because web search results are not ranked by relevance.
e)
Because it perfectly balances precision and recall.
32.
What is the primary goal of the 'stemming' process in text preprocessing?
a)
To remove frequently occurring but low-meaning words from the text.
b)
To correct spelling errors in the document.
c)
To reduce different grammatical forms of a word to a common base or root form.
d)
To identify and tag parts of speech for each word.
e)
To convert all characters to a single case (e.g., lowercase).
33.
Which of the following is an example of a 'stop word' that would likely be removed during text preprocessing?
a)
retrieval
b)
computer
c)
algorithm
d)
the
e)
database
34.
What is a potential disadvantage of aggressive stemming, as illustrated by the example of 'cop' and 'cope' both reducing to 'cop'?
a)
It significantly increases the size of the vocabulary.
b)
It can decrease precision by conflating distinct concepts.
c)
It makes the inverted index more complex to build.
d)
It improves recall at the expense of storage space.
e)
It requires a large, manually curated dictionary of root words.
35.
During the preprocessing of web pages, why is text within HTML tags like `<h1>` or `<b>` often given a higher weight?
a)
Because this text is usually longer and more descriptive.
b)
To penalize pages that overuse HTML for formatting.
c)
Because this text is programmatically easier to extract.
d)
Because it is considered a strong indicator of the page's main topic.
e)
Because it is guaranteed to be free of stop words.
36.
Why is the 'anchor text' of a hyperlink pointing to a page considered particularly important for web search?
a)
It indicates the geographic location of the linked page.
b)
It provides an objective, third-party description of the linked page's content.
c)
It is usually written by the author of the linked page.
d)
It is guaranteed to be a unique identifier for the page.
e)
It helps determine the font size and color of the linked text.
37.
For detecting *near* or *partial* duplicate documents, which technique is more suitable than simple hashing?
a)
Comparing file sizes.
b)
Using a checksum algorithm like MD5.
c)
Representing documents using n-grams and calculating Jaccard similarity.
d)
Comparing the number of stop words in each document.
e)
Stemming all words and comparing the resulting vocabulary.
38.
What is the primary advantage of using an inverted index over a simple sequential scan of documents for query processing?
a)
It reduces the storage space required for the document collection.
b)
It significantly speeds up the process of finding documents that contain query terms.
c)
It automatically ranks documents by relevance without further calculation.
d)
It handles synonymy and polysemy.
e)
It eliminates the need for text preprocessing steps like stemming.
39.
In a standard inverted index, what information is typically stored in the 'postings list' for each term?
a)
A list of all other terms that co-occur with it.
b)
The tf-idf score of the term in the collection.
c)
A list of document identifiers (and optionally, positions) where the term appears.
d)
A compressed version of the documents containing the term.
e)
The semantic concept associated with the term.
40.
When processing a multi-word query using an inverted index, a common optimization is to start by processing the postings list of which term?
a)
The term with the longest postings list.
b)
The term that appears first in the query.
c)
The term with the shortest postings list.
d)
The term with the highest idf score.
e)
A randomly selected term from the query.
41.
What is the purpose of storing term positions within the postings list of an inverted index?
a)
To calculate the tf-idf score more accurately.
b)
To enable processing of phrase and proximity queries.
c)
To reduce the storage size of the index.
d)
To speed up the merging of multiple postings lists.
e)
To help with stop word removal.
42.
What is a major challenge when building an inverted index for a very large collection of documents like the web?
a)
The vocabulary size becomes too small.
b)
The index may not fit into the main memory (RAM) of a single machine.
c)
The time to stem each word becomes the main bottleneck.
d)
The number of stop words exceeds the number of content words.
e)
The positional information cannot be stored.
43.
One strategy for building a large inverted index with limited RAM is to create partial indexes and then combine them. What is this combination process called?
a)
Merging
b)
Smoothing
c)
Hashing
d)
Ranking
e)
Compressing
44.
In index compression, what is 'delta encoding' or 'gap encoding' used for?
a)
To compress the vocabulary (term dictionary).
b)
To store the differences between consecutive sorted document IDs instead of the full IDs.
c)
To represent term frequencies using fewer bits.
d)
To encode the positional information of terms.
e)
To apply a lossless compression algorithm like Gzip to the entire index.
45.
What is a key trade-off between variable-byte and variable-bit index compression methods?
a)
Variable-bit methods offer better compression but can be slower to decode than byte-aligned methods.
b)
Variable-byte methods cannot handle large document IDs.
c)
Variable-bit methods are lossless, while variable-byte methods are lossy.
d)
Variable-byte methods are only suitable for static collections.
e)
Variable-bit methods require more storage but are faster to build.
46.
What is the primary problem in information retrieval that Latent Semantic Indexing (LSI) aims to solve?
a)
Slow query processing times due to large indexes.
b)
The vocabulary mismatch problem caused by synonymy and polysemy.
c)
Inefficient storage of the inverted index.
d)
The inability of the vector space model to rank documents.
e)
The difficulty of removing stop words from queries.
47.
LSI is based on the assumption that there is an underlying or 'latent' semantic structure in word usage patterns. Which mathematical technique is used to uncover this structure?
a)
Fourier Transform
b)
Clustering
c)
Singular Value Decomposition (SVD)
d)
Gradient Descent
e)
Naive Bayes Classification
48.
In the SVD of a term-document matrix A (A = U * Sigma * V^T), which matrix represents the documents in the latent semantic space?
a)
U
b)
Sigma
c)
V
d)
A^T (the transpose of A)
e)
The inverse of Sigma
49.
After performing SVD, LSI reduces the dimensionality by keeping only the top 'k' singular values. What is the main purpose of this reduction?
a)
To make the matrices invertible.
b)
To reduce noise and capture the most important semantic associations.
c)
To convert all values to positive integers.
d)
To prepare the data for Boolean query processing.
e)
To increase the sparsity of the term-document matrix.
50.
How does LSI allow a query containing 'automobile' to match a document containing 'car', even if the query word is not in the document?
a)
By using a manually built thesaurus to expand the query.
b)
By projecting both the query and the document into a reduced-dimension semantic space where they are close.
c)
By forcing all documents to use a controlled vocabulary.
d)
By applying a more aggressive stemming algorithm.
e)
By using a Boolean OR operator between all synonyms.
51.
What is a major disadvantage of using LSI for large-scale systems like web search?
a)
It only works for the Boolean model.
b)
The high computational cost of performing SVD on a large term-document matrix.
c)
It cannot handle new documents being added to the collection.
d)
It tends to have lower recall than the standard vector space model.
e)
The resulting semantic space is difficult to interpret.
52.
What key element, in addition to content relevance, do modern web search engines use to rank pages?
a)
The age of the domain name.
b)
The number of images on the page.
c)
The link structure of the web (hyperlinks).
d)
The length of the page's HTML code.
e)
The server's geographic location.
53.
In the context of web search, the idea that a hyperlink from page A to page B is a vote of confidence from A to B is the fundamental concept behind which type of analysis?
a)
Content analysis
b)
Duplicate content detection
c)
Lexical analysis
d)
Link structure analysis
e)
Sentiment analysis
54.
When evaluating the content of a web page, which location of a keyword would likely be given the highest importance by a search engine?
a)
In the copyright notice at the footer.
b)
In an HTML comment tag.
c)
In the page's title tag.
d)
Repeated many times in a hidden text block.
e)
In the filename of an image on the page.
55.
What is the primary function of a meta-search engine?
a)
It crawls and indexes the web independently.
b)
It provides search results only for academic papers.
c)
It sends a user's query to multiple other search engines and aggregates the results.
d)
It specializes in searching a specific topic, like medical information.
e)
It is another name for a web directory like Yahoo!.
56.
Which of the following is a primary motivation for using a meta-search engine?
a)
To get faster query response times.
b)
To improve search coverage by combining results from different indexes.
c)
To avoid seeing advertisements in search results.
d)
To use a simpler query syntax.
e)
To build a local, private index of the web.
57.
When combining results from multiple search engines, what does a method like 'CombSUM' do?
a)
It takes the minimum relevance score for each document across all search engines.
b)
It multiplies the relevance score by the number of search engines that returned the document.
c)
It sums the relevance scores for each document across all search engines.
d)
It takes the maximum relevance score for each document across all search engines.
e)
It re-ranks documents based on their original rank positions.
58.
The Borda Count method for rank aggregation works by assigning what to each document in a ranked list?
a)
A binary score (relevant or not relevant).
b)
A probabilistic relevance score.
c)
A point value based on its rank position.
d)
A pairwise comparison score against every other document.
e)
A score equal to the inverse of its rank.
59.
What is the fundamental difference between the Borda Count and Condorcet methods for rank aggregation?
a)
Borda is a voting method based on points, while Condorcet is based on pairwise comparisons.
b)
Borda only considers the top-ranked document, while Condorcet considers all.
c)
Borda is used for score fusion, while Condorcet is used for rank fusion.
d)
Borda is computationally efficient, while Condorcet is NP-hard.
e)
Borda is deterministic, while Condorcet is probabilistic.
60.
In Reciprocal Rank fusion, a document ranked at position 'r' by a search engine receives a score of 1/r. What score does a document receive if it is not returned by that search engine?
a)
A score of -1
b)
A score of 0
c)
A score of 1/(n+1), where n is the number of returned documents
d)
A small, default penalty score
e)
The average score of all returned documents
61.
What is the primary goal of web spam (also known as spamdexing)?
a)
To improve the user experience of search engines.
b)
To provide users with highly relevant and authoritative information.
c)
To artificially inflate a web page's ranking in search results for financial or other gain.
d)
To test the security vulnerabilities of web crawlers.
e)
To reduce the load on search engine servers.
62.
Repeating important keywords in a document, sometimes in a way that is nonsensical to a human reader, is a form of content spam aimed at manipulating which factor?
a)
Inverse Document Frequency (idf)
b)
Term Frequency (tf)
c)
Document Length Normalization
d)
PageRank
e)
Anchor Text
63.
Adding a large block of popular but unrelated keywords to a webpage (e.g., adding celebrity names to a travel site) is an attempt to achieve what?
a)
To increase the site's authority on its actual topic.
b)
To make the page relevant for a wider range of queries.
c)
To decrease the page's loading time.
d)
To attract high-quality incoming links.
e)
To improve the grammatical correctness of the page.
64.
What is 'link spam'?
a)
The practice of creating broken hyperlinks on a page.
b)
The practice of manipulating the number or quality of hyperlinks pointing to a page to boost its rank.
c)
The practice of using excessively long anchor text for all links.
d)
The practice of linking only to pages within the same website.
e)
The practice of creating hyperlinks that download malicious software.
65.
Creating a 'honeypot' page with useful, high-quality content to attract legitimate incoming links, which then secretly links to a target page to be boosted, is a form of which type of spam?
a)
Content Spam
b)
Cloaking
c)
Link Spam
d)
URL Redirect Spam
e)
Keyword Stuffing
66.
Cloaking' is a web spam technique where a web server...
a)
Encrypts the content of the page.
b)
Returns different content to a search engine crawler than it does to a human user's browser.
c)
Redirects the user to a different page immediately.
d)
Hides text by making it the same color as the background.
e)
Removes all hyperlinks from the page.
67.
Which of the following is a content-based feature that might indicate a page is spam?
a)
The page has a very short title.
b)
The average word length is significantly different from natural language norms.
c)
The page uses a common font like Arial.
d)
The page contains very few hyperlinks.
e)
The page has a .com top-level domain.
68.
How does the analysis of anchor text from incoming links help combat web spam?
a)
Spam pages rarely have incoming links with anchor text.
b)
Anchor text provides an external, often more objective description of a page's content, which is harder for spammers to manipulate directly.
c)
It allows search engines to identify the IP address of the spammer.
d)
It increases the weight of keywords found within the spam page itself.
e)
It is used to calculate the tf-idf score of the spam page.
69.
Which core component of an IR system is built by processing a collection of raw documents to create a searchable data structure?
a)
The user interface
b)
The query parser
c)
The text index
d)
The relevance feedback module
e)
The result display
70.
If a user query is 'the best smart phone', what is a likely first step in query processing?
a)
Calculating the tf-idf scores for 'best' and 'smart'.
b)
Searching for the exact phrase 'the best smart phone'.
c)
Removing the stop word 'the'.
d)
Expanding 'smart phone' with synonyms like 'cell phone'.
e)
Translating the query into a different language.
71.
In the Vector Space Model, documents and queries are represented as:
a)
Strings of text.
b)
Sets of keywords.
c)
Probabilistic language models.
d)
Vectors in a high-dimensional space.
e)
Nodes in a graph.
72.
What does a high IDF score for a term imply?
a)
The term is very common in the document collection.
b)
The term is a stop word.
c)
The term is rare in the document collection and thus more discriminative.
d)
The term appears frequently within a specific document.
e)
The document containing the term is very long.
73.
The Okapi BM25 formula includes document length (`dl_j`) and average document length (`avdl`). What is the purpose of including these terms?
a)
To give a higher score to longer documents, as they are more comprehensive.
b)
To penalize documents that are much longer or shorter than the average, normalizing for length.
c)
To calculate the cosine similarity between the document and the query.
d)
To ensure the final score is between 0 and 1.
e)
To estimate the probability of the document generating the query.
74.
In a statistical language model, if a query word does not appear in a document, what would its unsmoothed probability Pr(word|document) be?
a)
A very small positive number.
b)
1
c)
0.5
d)
A negative number.
75.
In the Rocchio formula for relevance feedback, what would happen if the gamma parameter (weight for non-relevant documents) was set to zero?
a)
The query would not be modified.
b)
The query would only be moved away from non-relevant documents.
c)
The expanded query would only be influenced by the original query and the relevant documents.
d)
The influence of the original query would be removed.
e)
Only non-relevant documents would be used for expansion.
76.
Which evaluation metric focuses on the fraction of retrieved documents that are relevant?
a)
Recall
b)
F-Score
c)
Precision
d)
Average Precision
e)
Mean Average Precision
77.
If a search returns 5 documents, and the 1st, 3rd, and 5th are relevant, what is the precision at position 3 (P@3)?
a)
1/3
b)
2/3
c)
3/3
d)
2/5
e)
3/5
78.
If a search returns 5 documents, of which 3 are relevant, but there are a total of 10 relevant documents in the entire collection, what is the recall of this search?
a)
3/5 (60%)
b)
5/10 (50%)
c)
3/10 (30%)
d)
5/3
e)
10/3
79.
Why might a search engine developer prefer the F-score over just precision or just recall?
a)
It is easier to calculate.
b)
It provides a single measure that balances the trade-off between precision and recall.
c)
It is specifically designed for evaluating web search.
d)
It ignores the ranking of the documents.
e)
It is a measure of system efficiency, not effectiveness.
80.
Which text processing step would convert 'computing', 'computed', and 'computation' to the common form 'comput'?
a)
Stop word removal
b)
Stemming
c)
Tokenization
d)
Case folding
e)
Lemmatization
81.
In preprocessing web pages, why is it important to identify the main content blocks?
a)
To remove all content except for the navigation bar and footer.
b)
To apply different weighting schemes to main content versus advertisements or boilerplate text.
c)
To ensure the page can be correctly rendered in a browser.
d)
To extract all the email addresses present on the page.
e)
To speed up the download time of the page.
82.
To detect if `doc_A` is an exact duplicate of `doc_B`, the most efficient method would be to:
a)
Compare their n-gram sets.
b)
Calculate the cosine similarity of their tf-idf vectors.
c)
Compare their checksums or hash values.
d)
Count the number of sentences in each.
e)
Compare their file creation dates.
83.
How is a phrase query like 'web mining' processed using an inverted index with positional information?
a)
By finding documents containing 'web' OR 'mining'.
b)
By finding documents that contain both 'web' AND 'mining', and then checking if 'mining' appears immediately after 'web'.
c)
By merging the postings lists and summing the term frequencies.
d)
By calculating the cosine similarity for 'web' and 'mining' separately.
e)
By creating a new index specifically for two-word phrases.
84.
A key benefit of using a Trie or B-Tree to store the vocabulary in an inverted index is that it:
a)
Reduces the storage needed for postings lists.
b)
Allows for fast lookup of terms in the vocabulary.
c)
Handles stop word removal automatically.
d)
Compresses the document collection.
e)
Pre-calculates all relevance scores.
85.
What does Elias gamma coding, a variable-bit compression technique, do particularly well?
a)
It is very efficient at encoding large numbers.
b)
It is very efficient at encoding small positive integers.
c)
It is a byte-aligned code, making it fast to decode.
d)
It is a form of lossy compression.
e)
It is used to compress the vocabulary, not the postings lists.
86.
How does LSI's use of SVD help address the problem of synonymy?
a)
It replaces all words with their most common synonym from a thesaurus.
b)
It maps semantically similar words (like 'boat' and 'ship') to similar vectors in the latent concept space.
c)
It removes all ambiguous words from the vocabulary.
d)
It forces documents to be clustered based on keyword overlap only.
e)
It increases the weight of rare words, which are less likely to be synonyms.
87.
In the SVD decomposition `A = U * Sigma * V^T`, what do the diagonal values in the `Sigma` matrix represent?
a)
The relevance of each document.
b)
The importance or strength of each latent concept (dimension).
c)
The frequency of each term.
d)
The mapping from terms to concepts.
e)
The mapping from documents to concepts.
88.
What is a major conceptual difference between traditional keyword-based IR and LSI?
a)
LSI does not require an index.
b)
LSI matches queries and documents in a concept space rather than a term space.
c)
LSI is faster for web-scale collections.
d)
LSI can only be used for Boolean queries.
e)
LSI does not use term weighting.
89.
A search for 'human computer interaction' is performed. Why might an LSI-based system return a document titled 'A Study of User Interfaces for HCI', even if the query words are not present?
a)
Because of a lucky random guess.
b)
Because 'user' and 'interface' are highly correlated with the query terms in the collection's latent semantic space.
c)
Because the document has a high PageRank score.
d)
Because the document title is very short.
e)
Because 'HCI' is a stop word.
90.
The process of a web crawler visiting web pages, parsing their content, and following links to other pages is the first step in:
a)
Query processing.
b)
Building a search engine's index.
c)
Relevance feedback.
d)
Web spam detection.
e)
Calculating PageRank.
91.
A meta-search engine receives ranked lists from three different search engines. To combine them, it gives 3 points for a 1st place rank, 2 for 2nd, and 1 for 3rd. It then sums the points for each URL. This is an example of which fusion method?
a)
Condorcet Fusion
b)
Reciprocal Ranking
c)
CombSUM
d)
Borda Count
e)
CombMIN
92.
What is the defining characteristic of 'keyword stuffing' as a form of content spam?
a)
Creating many pages with slightly different keywords.
b)
Hiding keywords using scripts or tiny fonts.
c)
Loading a page with a high density of certain keywords to manipulate term frequency.
d)
Placing keywords in the URL of the page.
e)
Using keywords that are completely unrelated to the page's visible content.
93.
A spammer creates a network of websites that all link to each other and to a central target website. This technique, designed to manipulate link-based ranking algorithms, is known as a:
a)
Honeypot
b)
Link farm
c)
Cloaking network
d)
Mirror site
e)
Dead end page
94.
If a web server detects the IP address of a known search engine crawler and serves it a page full of keywords, but serves a normal-looking page to all other visitors, this is an example of:
a)
Keyword stuffing
b)
Link spam
c)
Doorway pages
d)
Cloaking
e)
Page hijacking
95.
Why is detecting and penalizing web spam crucial for a search engine?
a)
It reduces the size of the index.
b)
It improves the relevance of search results and maintains user trust.
c)
It increases the number of advertisements that can be shown.
d)
It helps crawlers discover new pages more quickly.
e)
It is required by international law.
96.
Which of the following is a characteristic of the Boolean information retrieval model?
a)
It provides a ranked list of documents based on probable relevance.
b)
It represents documents as vectors in a multi-dimensional space.
c)
It processes queries based on exact set theory and logical operators.
d)
It uses term frequency to weigh the importance of keywords.
e)
It is designed to handle natural language queries effectively.
97.
In the tf-idf weighting scheme, if a term appears in every single document in the collection, what will its idf score be?
a)
Very high, because it is a common term.
b)
Zero, because log(N/N) = log(1) = 0.
c)
Exactly 1.
d)
A negative number.
e)
Dependent on the term's frequency within each document.
98.
What is the primary motivation for using the Rocchio algorithm?
a)
To build an inverted index more quickly.
b)
To automatically reformulate a query based on user feedback to improve results.
c)
To compress the document collection without losing information.
d)
To find the principal components of the term-document matrix.
e)
To rank documents using a probabilistic model.
99.
If an IR system retrieves 10 documents, of which 6 are relevant, but the total number of relevant documents in the collection is 20, what are the precision and recall?
a)
Precision = 6/10, Recall = 6/20
b)
Precision = 6/20, Recall = 6/10
c)
Precision = 10/6, Recall = 20/6
d)
Precision = 6/10, Recall = 10/20
e)
Precision = 4/10, Recall = 14/20
100.
A document preprocessing pipeline consists of: 1. Case folding, 2. Stop word removal, 3. Stemming. Why is this order logical?
a)
Stemming must be done first to create a root word dictionary.
b)
The order does not matter; the final result will be the same.
c)
Case folding is done first so that stop words like 'The' and 'the' are both recognized and removed before stemming.
d)
Stop words must be removed first, otherwise the stemmer will not work correctly.
e)
Case folding is the most computationally expensive and should be last.
101.
What is the main purpose of storing 'offsets' or 'positions' in an inverted index's postings list?
a)
To calculate the document length without accessing the original document.
b)
To allow for the processing of phrase queries and proximity queries.
c)
To speed up the process of merging two postings lists for an AND query.
d)
To reduce the storage space of the index through gap encoding.
e)
To assign a unique ID to each term occurrence.
102.
What is the core idea behind index compression techniques like Elias codes or Golomb codes?
a)
To use fewer bits to represent more frequent (i.e., smaller) numbers, such as the gaps between document IDs.
b)
To apply a standard compression algorithm like ZIP to the entire index file.
c)
To remove terms from the index that are unlikely to be queried.
d)
To store the index in a byte-aligned format for faster CPU processing.
e)
To create a lossy representation of the index to save space.
103.
Which statement best describes the role of the 'k' parameter (number of dimensions) in Latent Semantic Indexing?
a)
A higher 'k' always results in better retrieval performance.
b)
A very low 'k' may over-generalize, while a very high 'k' may fail to capture synonyms and just models noise.
c)
k' must be equal to the number of documents in the collection.
d)
k' is determined by the number of terms in the query.
e)
The optimal value of 'k' is always 2 or 3 for visualization.
104.
In web search, what does PageRank primarily measure?
a)
The relevance of a page's content to a query.
b)
The number of keywords on a page.
c)
The 'importance' or 'authority' of a page based on the web's link structure.
d)
The age of the web page.
e)
The speed at which the page loads.
105.
If a meta-search engine uses the Condorcet method, a document A is ranked higher than document B if:
a)
A has a higher cumulative Borda score than B.
b)
A is returned by more search engines than B.
c)
A is ranked higher than B by a majority of the individual search engines.
d)
A has a higher average relevance score than B.
e)
A's rank is the reciprocal of B's rank.
106.
A webpage contains a long list of city names in text that is the same color as the page background. This is an example of:
a)
A link farm.
b)
A doorway page.
c)
Content spam using hidden text.
d)
Cloaking
e)
A honeypot.
107.
TrustRank is an algorithm used to combat web spam. What is its core principle?
a)
It assumes that pages with more content are more trustworthy.
b)
It starts with a seed set of trusted pages and propagates trust through links, with less trust propagating from spammy pages.
c)
It trusts pages more if they have a .edu or .gov domain.
d)
It trusts pages that do not link to any external websites.
e)
It assigns trust based on the age of the website.
108.
What fundamental assumption of the 'bag of words' model makes it difficult to distinguish between 'dog bites man' and 'man bites dog'?
a)
It ignores word frequency.
b)
It requires stemming.
c)
It disregards word order and proximity.
d)
It cannot handle a large vocabulary.
e)
It represents terms with binary weights.
109.
Why is smoothing essential for statistical language models in IR?
a)
It makes the model run faster.
b)
It prevents the probability of an entire query from becoming zero just because one of its terms is missing from a document.
c)
It normalizes all document vectors to unit length.
d)
It helps to identify stop words.
e)
It selects the best language model (e.g., unigram vs. bigram).
Reset
