cisc 7610 lecture 7 principles of content-based...

CISC 7610 Lecture 7Principles of content-based retrieval

Topics:Housekeeping

Content-based retrievalInformation retrieval

Term-document matricesInverted indices

Boolean queryingEvaluating retrieval

Housekeeping

● Please complete the midterm feedback form I am distributing right now

Housekeeping

● Please complete the midterm feedback form I am distributing right now

● The midterm exam is graded– Min: 63

– Mean: 79

– Median: 80

– Max: 93

● Let's go over the solutions

Course project

● Half of your final grade is based on the final project, implementing a multimedia database

● Your idea does not have to be totally new, but your implementation should be original

● You will be graded on three components– Project proposal

– Project presentation

– Project writeup and code submission

● You may work in groups

Course project: Proposal

● The project proposal should be a 2-page writeup

● It should describe– What problem are you addressing?

– What multimedia data will you index?

– What sort of queries will your system support?

– What technologies will you use to address it?

– You may want to include some sort of system diagram sketch or other representation of the various system components and how they fit together

Course project examples

● Image database that indexes images using facial recognition software and allows query-by-image

● Image database that classifies images according to the objects they contain and allows search-by-description (see Caffe image classifier from Berkeley)

● Video database that allows closed captions to be searched and skips to those locations in the video

● Speech database that recognizes what is said and allows keyword search, skipping to the relevant locations

● Video database that identifies text on-screen and allows it to be searched

Recall: Main question

How can we process and store multimedia data so that we can

find what we are looking for in the future?

Information Retrieval

● Text Information Retrieval (IR) is finding material (usually documents) of an unstructured nature (usually text) that satisfies an information need from within large collections (usually stored on computers)

● Multimedia Information Retrieval (MIR) is finding material of a multimedia nature that satisfies an information need from within large collections

● Multimedia IR has a lot in common with text IR, especially when moving beyond SQL

8

Content-based retrieval

● Consider using an image as a query:

● Q1: Tell me about this building

● Q2: Find other buildings like this one

● Q3: Find paintings and drawings of this building

http://www.brooklyn.cuny.edu/web/library.php

Content-based retrieval

● Consider using an image as a query:

● These queries can best be understood and answered using ideas from multimedia information retrieval

http://www.brooklyn.cuny.edu/web/library.php

Similar text retrieval

● Similar queries would occur for text retrieval:– Q1: Find me web pages about “Brooklyn College”

– Q2: Find other pages like http://www.brooklyn.cuny.edu

– Q3: Find books that mention “Brooklyn College”

● Text information retrieval has been extensively studies for decades

http://www.brooklyn.cuny.edu/

Basic assumptions of Information Retrieval

• Collection: A set of documents– Assume it is a static collection for the moment

• Goal: Retrieve documents with information that is relevant to the user’s information need and helps the user complete a task

12

Sec. 1.1

Manning and Nayak, CS276, Lecture 1: Introduction http://web.stanford.edu/class/cs276/

http://web.stanford.edu/class/cs276/


how trap mice alive

The classic search model

Collection

User task

Info need

Query

Results

Searchengine

Queryrefinement

Get rid of mice in a politically correct way

Info about removing micewithout killing them

Misconception?

Misformulation?

Search


How good are the retrieved docs?

Precision : Fraction of retrieved docs that are relevant to the user’s information need

Recall : Fraction of relevant docs in collection that are retrieved

More precise definitions and measurements later

14

Sec. 1.1



Term-document incidence matrices

Unstructured data in 1620

● Which plays of Shakespeare contain the words Brutus AND Caesar but NOT Calpurnia?

● One could grep all of Shakespeare’s plays for Brutus and Caesar, then strip out lines containing Calpurnia?

● Why is that not the answer?– Slow (for large corpora)– NOT Calpurnia is non-trivial– Other operations (e.g., find the word Romans near

countrymen) not feasible– Ranked retrieval (best documents to return)

16

Sec. 1.1



Term-document incidence matrices

Antony and Cleopatra J ulius Caesar The Tempest Hamlet Othello Macbeth

Antony 1 1 0 0 0 1

Brutus 1 1 0 1 0 0

Caesar 1 1 0 1 1 1

Calpurnia 0 1 0 0 0 0

Cleopatra 1 0 0 0 0 0

mercy 1 0 1 1 1 1

worser 1 0 1 1 1 0

1 if play contains word, 0 otherwise

Brutus AND Caesar BUT NOT Calpurnia

Sec. 1.1



Incidence vectors

● So we have a 0/1 vector for each term.● To answer query: take the vectors for Brutus,

Caesar and Calpurnia (complemented) bitwise AND.

110100 AND110111 AND101111 =

100100

18

Sec. 1.1

Antony and Cleopatra J ulius Caesar The Tempest Hamlet Othello Macbeth

Antony 1 1 0 0 0 1

Brutus 1 1 0 1 0 0

Caesar 1 1 0 1 1 1

Calpurnia 0 1 0 0 0 0

Cleopatra 1 0 0 0 0 0

mercy 1 0 1 1 1 1

worser 1 0 1 1 1 0



Answers to query

● Antony and Cleopatra, Act III, Scene iiAgrippa [Aside to DOMITIUS ENOBARBUS]: Why, Enobarbus,

When Antony found Julius Caesar dead,

He cried almost to roaring; and he wept

When at Philippi he found Brutus slain.

● Hamlet, Act III, Scene iiLord Polonius: I did enact Julius Caesar I was killed i’ the

Capitol; Brutus killed me.

19

Sec. 1.1



Bigger collections

● Consider N = 1 million documents, each with about 1000 words.

● Avg 6 bytes/word including spaces/punctuation – 6GB of data in the documents.

● Say there are M = 500K distinct terms among these.

20

Sec. 1.1



Can’t build the matrix

● 500K x 1M matrix has half-a-trillion 0’s and 1’s.

● But it has no more than one billion 1’s.– matrix is extremely sparse.

● What’s a better representation?– We only record the 1 positions.

21

Why?

Sec. 1.1



The inverted index:

They key data structure underlying modern IR

Inverted index

● For each term t, we must store a list of all documents that contain t.– Identify each doc by a docID, a document serial

number● Can we used fixed-size arrays for this?

23

What happens if the word Caesar is added to document 14?

Sec. 1.2

Brutus

Calpurnia

Caesar 1 2 4 5 6 16 57 132

1 2 4 11 31 45 173

2 31

174

54 101




Inverted index

● We need variable-size postings lists– On disk, a continuous run of postings is best– In memory, can use linked lists or variable length

arrays● Some tradeoffs in size/ease of insertion

24

Dictionary Postings

Sorted by docID (more later on why).

PostingPosting

Sec. 1.2

Brutus

Calpurnia

Caesar 1 2 4 5 6 16 57 132

1 2 4 11 31 45 173

2 31

174

54 101



Indexer steps: Token sequence

● Sequence of (Modified token, Document ID) pairs.

I did enact JuliusCaesar I was killed

i’ the Capitol; Brutus killed me.

Doc 1

So let it be withCaesar. The noble

Brutus hath told youCaesar was ambitious

Doc 2

Sec. 1.2



Indexer steps: Sort

● Sort by terms– And then docID

Core indexing step

Sec. 1.2



Indexer steps: Dictionary & Postings

● Multiple term entries in a single document are merged.

● Split into Dictionary and Postings

● Doc. frequency information is added.

Why frequency?Will discuss later.

Sec. 1.2



Where do we pay in storage?

28Pointers

Terms and

counts IR system implementation•How do we index efficiently?•How much storage do we need?

Sec. 1.2

Lists of docIDs


Query processing with an inverted index

Query processing: AND

● Consider processing the query:Brutus AND Caesar– Locate Brutus in the Dictionary;

● Retrieve its postings.

– Locate Caesar in the Dictionary;● Retrieve its postings.

– “Merge” the two postings (intersect the document sets):

30

128

34

2 4 8 16 32 64

1 2 3 5 8 13 21

Brutus

Caesar

Sec. 1.3



The merge

● Walk through the two postings simultaneously, in time linear in the total number of postings entries

31

34

1282 4 8 16 32 64

1 2 3 5 8 13 21

Brutus

Caesar

If the list lengths are x and y, the merge takes O(x+y)operations.Crucial: postings sorted by docID.

Sec. 1.3



Intersecting two postings lists(a “merge” algorithm)

32Manning and Nayak, CS276, Lecture 1: Introduction http://web.stanford.edu/class/cs276/


Boolean queries: Exact match

● The Boolean retrieval model is being able to ask a query that is a Boolean expression:– Boolean Queries are queries using AND, OR and NOT to join

query terms● Views each document as a set of words● Is precise: document matches condition or not.

– Perhaps the simplest model to build an IR system on● Primary commercial retrieval tool for 3 decades. ● Many search systems you still use are Boolean:

– Email, library catalog, Mac OS X Spotlight

33

Sec. 1.3



Boolean queries: More general merges

• Exercise: Adapt the merge for the queries:

Brutus AND NOT Caesar

Brutus OR NOT Caesar

● Can we still run through the merge in time O(x+y)? What can we achieve?

34

Sec. 1.3



Merging

What about an arbitrary Boolean formula?

(Brutus OR Caesar) AND NOT(Antony OR Cleopatra)● Can we always merge in “linear” time?

– Linear in what?● Can we do better?

35

Sec. 1.3



Query optimization

● What is the best order for query processing?● Consider a query that is an AND of n terms.● For each of the n terms, get its postings, then

AND them together.

Brutus

Caesar

Calpurnia

1 2 3 5 8 16 21 34

2 4 8 16 32 64 128

13 16

Query: Brutus AND Calpurnia AND Caesar36

Sec. 1.3



Query optimization example

● Process in order of increasing freq:– start with smallest set, then keep cutting further.

37

This is why we keptdocument freq. in dictionary

Execute the query as (Calpurnia AND Brutus) AND Caesar.

Sec. 1.3

Brutus

Caesar

Calpurnia

1 2 3 5 8 16 21 34

2 4 8 16 32 64 128

13 16



More general optimization

● e.g., (madding OR crowd) AND (ignoble OR strife)● Get doc. freq.’s for all terms.● Estimate the size of each OR by the sum of its doc.

freq.’s (conservative).● Process in increasing order of OR sizes.

38

Sec. 1.3



Exercise• Recommend a query processing

order for

• Which two terms should we process first?

Term Freq eyes 213312 kaleidoscope 87009 marmalade 107913 skies 271658 tangerine 46653 trees 316812

39

(tangerine OR trees) AND(marmalade OR skies) AND(kaleidoscope OR eyes)



Query processing exercises

• Exercise: If the query is friends AND romans AND (NOT countrymen), how could we use the freq of countrymen?

• Exercise: Extend the merge to an arbitrary Boolean query. Can we always guarantee execution in time linear in the total postings size?

• Hint: Begin with the case of a Boolean formula query: in this, each query term appears only once in the query.

40Manning and Nayak, CS276, Lecture 1: Introduction http://web.stanford.edu/class/cs276/


Evaluating retrieval

42

● How fast does it index Number of documents/hour (Average document size)

● How fast does it search Latency as a function of index size

● Expressiveness of query language Ability to express complex information needs Speed on complex queries

● Uncluttered UI● Is it free?

Sec. 8.6

Measures for a search engine

Manning and Nayak, CS276, Lecture 8: Evaluation http://web.stanford.edu/class/cs276/


43

Most important: user happiness But difficult to measure

Most common proxy: relevance of search results Relevance measurement requires 3 elements:

1. A benchmark document collection

2. A benchmark suite of queries

3. A usually binary assessment of either Relevant or Nonrelevant for each query and each document Some work on more-than-binary, but not the standard

Sec. 8.1

Measures for a search engine



44

● Note: the information need is translated into a query● Relevance is assessed relative to the information need

not the query● E.g., Information need: I'm looking for information on

whether drinking red wine is more effective at reducing your risk of heart attacks than white wine.

● Query: wine red white heart attack effective● Evaluate whether the doc addresses the information

need, not whether it has these words

Sec. 8.1

Evaluating an IR system



Example: Image retrieval

Rueger, “Multimedia Information Retrieval” Figure 2.11. Morgan & Claypool: 2010.

48

Unranked retrieval evaluation:Precision and recall

● Precision: fraction of retrieved docs that are relevant = P(relevant|retrieved)

● Recall: fraction of relevant docs that are retrieved= P(retrieved|relevant)

● Precision P = tp/(tp + fp)● Recall R = tp/(tp + fn)

Relevant Nonrelevant

Retrieved True positives False positives

Not Retrieved False negatives True negatives

Sec. 8.3



49

Should we instead use the accuracy measure for evaluation?

● Given a query, an engine classifies each doc as “Relevant” or “Nonrelevant”

● The accuracy of an engine: the fraction of these classifications that are correct (tp + tn) / ( tp + fp + fn + tn)

● Accuracy is a commonly used evaluation measure in machine learning classification work

● Why is this not a very useful evaluation measure in IR?

Sec. 8.3



50

Search for:

0 matching results found.

Sec. 8.3

Why not just use accuracy?

● How to build a 99.9999% accurate search engine on a low budget….

● People doing information retrieval want to find something and have a certain tolerance for junk.



Precision / recall example

● We have a database with 10,000 documents in it– 8 are relevant to the current query

● The IR system returns 20 results:– R = relevant, N = not relevant

● What the precision of the system?– What is the precision-at-5? Precision-at-2?

● What is the recall of the system?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R R N N N N N N R N R N N N R N N N N R



52

● You can get high recall (but low precision) by retrieving all docs for all queries!

● Recall is a non-decreasing function of the number of docs retrieved

● In a good system, precision decreases as either the number of docs retrieved or recall increases This is not a theorem, but a result with strong empirical

confirmation

Sec. 8.3

Precision/Recall



53

● Customers typically want a single figure of merit● Combined measure that assesses precision/recall tradeoff is F measure

(weighted harmonic mean):

● People usually use balanced F1 measure i.e., with = 1 or = ½

● Harmonic mean is a conservative average See CJ van Rijsbergen, Information Retrieval

RP

PR

RP

F

2

2 )1(1

)1(1

1

Sec. 8.3

A combined measure: F



Manning and Nayak, CS276, Lecture 8: Evaluation http://web.stanford.edu/class/cs276/ 54

● Instead of returning only some results, the system returns all results● But ordered from best to worst● The evaluator can then decide how many to keep● Leads to precision-recall curves

Sec. 8.4

Evaluating ranked results


Manning and Nayak, CS276, Lecture 8: Evaluation http://web.stanford.edu/class/cs276/ 55

● Idea: If locally precision increases with increasing recall, then you should get to count that…

● So take the max of precisions to right of value

Sec. 8.4

Interpolated precision


Precision / recall example

● We have a database with 10,000 documents in it– 8 are relevant to the current query

● The IR system returns 20 results:– R = relevant, N = not relevant

● What the F-measure of the system?

● What is the uninterpolated precision at 25% recall?

● What is the interpolated precision at 33% recall?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R R N N N N N N R N R N N N R N N N N R



● Search engines have test collections of queries and hand-ranked results● Recall is difficult to measure on the web● Search engines often use precision at top k, e.g., k = 10● . . . or measures that reward you more for getting rank 1 right than for

getting rank 10 right. NDCG (Normalized Cumulative Discounted Gain)

● Search engines also use non-relevance-based measures. Clickthrough on first result

Not very reliable if you look at a single clickthrough … but pretty reliable in the aggregate.

Studies of user behavior in the lab A/B testing

57

Sec. 8.6.3

Evaluation at large search engines



cisc 7610 lecture 7 principles of content-based...

Documents