Exam Deck

Question 1

Basic Measures of TR system and formulas

Answer 1

Precision = a / a+c - are retreived results relevant
Recall = a / a+b - have all relevant documents been retrieved
F-measure = 2PR/P+R - combines Precision and recall

Question 2

What is the ideal PR curve, what does it characterize

Answer 2

1. Horizontal line, Precision = recall

2. characterizes overall accuracy

Question 3

What is average precision

Answer 3

standard measure of comparing two ranking methods

• Combines Precision and recall
• Sensitive to rank of every relevant document

Question 4

What is nDCG

Answer 4

• utility of top k documents
• utility of lowly ranked document is discounted
• normalized across queries

Question 5

Describe all types of feedback

Answer 5

Relevance feedback - Reliable judgement, but requires effort

Pseudo Feedback - Not reliable, assumed top k ranked docs are reliable with no user effort

Implicit Feedback - Uses clickthrough

Question 6

What is Latent semantic indexing

Answer 6

• find a way to represent the term-document space by a lower dimension latent space
• improve storage and ambiguity search

Question 7

LSI steps

Answer 7

• Term document matrix -> Word assignment to topics -> Topic importance -> topic distribution

Question 8

Pros of using VSM

Answer 8

• Automatic selection of index terms
• Partial Matching of queries and documents
• Ranking to the similarity score
• Term weighting schemes
• Various extensions

Question 9

Problems with Lexical Semantics

Answer 9

• Synonymy = Different terms may have identical or similar meanings similarity is high even though cosine small
• Polysemy = words often have many meanings , vsm unable to discriminate. cosine large but should be small

Question 10

Advantages of Lexical

Question 11

Main idea of LSI

Answer 11

Perform a low-rank appx of the document term matrix
General Idea
- to map documents to low dimensions.
- represent semantic associations
- compute similarity based on the inner product in semantic space

Question 12

Web search challenges and opportunity

Answer 12

Challenges

• Scalability = Parallel indexing and searching
• Low-quality information and spam
• Dynamics of web

Opportunities
- many additional heuristics can be leveraged to improve accuracy

Question 13

What is web crawler

Answer 13

• an essential component of web search
• BFS
• complete vs focused crawling
• Incremental crawling is resource optimization

Question 14

what is mapreduce

Answer 14

• Minimizes effort of programmer for simple parallel programming tasks

Question 15

what is page rank algorithm

Answer 15

• captures page popularity

- random surfing to visit every page and assess the popularity

Question 16

What is hits algorithm

Answer 16

• pages with good authorities/ good hubs