Recap through exercise --

X x1 : {Block, Jeevan, Cinema, Old, Agra}

x2 : {House, Jeevan, Vihar, Delhi}

Y x3 : {Block, Jeevan, Cinema, Agra}

x4 : {Building, Ravi, Vihar, New, Delhi}
x5 : {Block, Jeevan, Vihar, New, Delhi}
x6 : {Mandir, Ravi, Vihar, Delhi}
Inverted Index

X x1 : {Block, Jeevan, Cinema, Old, Agra}

x2 : {House, Jeevan, Vihar, Delhi}
x1: {Block, Jeevan, Cinema, Old, Agra} Y x3 : {Block, Jeevan, Cinema, Agra}
x4 : {Building, Ravi, Vihar, New, Delhi}
x5 : {Block, Jeevan, Vihar, New, Delhi}
What will be the candidate set? x6 : {Mandir, Ravi, Vihar, Delhi}

IDX(S)
Candidate set: {3, 5}+{3, 5}+{3}+{3} Token DID
Agra 3
Block 3, 5

ID Freq Cinema 3
Delhi 4, 5, 6
3 4
Jeevan 3, 5
5 2
Mandir 6
New 4, 5
Ravi 4, 6
Vihar 4, 5, 6 2
 X x1 : {Block, Jeevan, Cinema, Old, Agra}
x2 : {House, Jeevan, Vihar, Delhi}
Size Filtering Y x3 : {Block, Jeevan, Cinema, Agra}
x4 : {Building, Ravi, Vihar, New, Delhi}
x5 : {Block, Jeevan, Vihar, New, Delhi}

Given a string xεX, only strings y such that |x|/t ≥ |y| ≥ |x|*t
x6 : {Mandir, Ravi, Vihar, Delhi}

Consider x1={Block, Jeevan, Cinema, Old, Agra}, suppose t=0.9

What will be the candidate set?

IDX(S)
• If yεY matches x, we must have Token DID
Agra 3
• 5/0.9 = 5.5 ≥ |y| ≥ 5* 0.9 = 4.5 Block 3, 4, 5
Cinema 3
• Candidate set C= x4, x5
Delhi 4, 5, 6
Jeevan 3, 5
Mandir 6
New 4, 5
Ravi 4, 6
3
Vihar 4, 5, 6
 Prefix Filtering

• if |x ⋂ y| ≥ k  any subset x’  x of size at least |x| - (k – 1)

must overlap y
• Consider matching using O(x,y) ≥ 3
X x : {Block, Jeevan, Cinema, Old, Agra}
•
1
x1={Block, Jeevan, Cinema, Old, Agra} x : {House, Jeevan, Vihar, Delhi}
2

Y x3 : {Block, Jeevan, Cinema, Agra}

x4 : {Building, Ravi, Vihar, New, Delhi}
x5 : {Block, Jeevan, Vihar, New, Delhi}
• What would be the candidate set? x6 : {Mandir, Ravi, Vihar, Delhi}

• Prefix x1’ = {Block, Jeevan, Cinema},

• Use inverted index to find the candidate set C={x3, x5} which
contain at least one token in x1’.

4
 Prefix-Inverted Index

Token ID List TF Order

X x1 : {Block, Jeevan, Cinema, Old, Agra}
x2 : {House, Jeevan, Vihar, Delhi} Agra 1, 3 2 5
Block 1, 3, 5 3 9
Y x3 : {Block, Jeevan, Cinema, Agra}
x4 : {Building, Ravi, Vihar, New, Delhi} Building 4 1 1
x5 : {Block, Jeevan, Vihar, New, Delhi}
Cinema 1, 3 2 6
x6 : {Mandir, Ravi, Vihar, Delhi}
Delhi 2, 4, 5, 6 4 10
House 2 1 2
Jeevan 1, 2, 3, 5 4 11
Order: Mandir 6 1 3
Building < House < Mandir < Old < Agra < Cinema
< New < Ravi < Block < Delhi < Jeevan < Vihar New 4, 5 2 7
Old 1 1 4
Ravi 4, 6 2 8
Vihar 2, 4, 5, 6 4 12
Order: x1 : {Block, Jeevan, Cinema, Old, Agra}
Building < House < Mandir < Old < Agra < Cinema < New < Ravi < Block < Delhi < Jeevan < Vihar

Algorithm
• reorder terms in each x ε X and y ε Y in increasing order of their frequencies
• for each y εY, create y’, the prefix of size |y| - (k – 1) of y
• build an inverted index over all prefixes y’
• for each x ε X, create x’, the prefix of size |x| - (k – 1) of x, then use above index
to find all y such that x’ overlaps with y’
Prefix Inverted Index for Y
O(x,y) ≥ 3 Prefix(x)=|x|-(3-1)=|x|-2 Token DID
Agra 3
What would be the candidate set?
X x1 : {Old, Agra, Cinema, Block, Jeevan} Cinema 3
x2 : {House, Delhi, Jeevan, Vihar} Delhi 5

Y x3 : {Cinema, Agra, Block, Jeevan} Mandir 6

x4 : {Building, New, Ravi, Delhi, Vihar} New 4, 5
x5 : {New, Block, Delhi, Jeevan, Vihar,}
Ravi 4, 6
x6 : {Mandir, Ravi, Delhi, Vihar}

Candidate Set C for x1={Old, Agra, Cinema, Block, Jeevan} is {x3}. 6

Position Index

x1={Old, Agra, Cinema, Block, Jeevan}

x3 : {Cinema, Agra, Block, Jeevan}
O(x,y) > 4

Estimation of max overlap = overlap in prefixes + min of

unseen tokens = 2 + min(2, 2) = 4 > k (false)
 No need to compute sim(x1, x3) !

7
More on Schema Matching and Mapping

8
Recap --
 creating semantic matches between schemas
 name = title, location = concat(city, state, zipcode)
 Defining semantic mapping between schemas using SQL

 Remember -- Schema matching and mapping are often

quite difficult

9
Schemas

BOOK-VENDOR
Books(ISBN, title, category, PubYear)
Shop(SID, BID, markedPrice, Vloc)

10
Semantic Matches

 A semantic match relates a set of elements in a schema S

to a set of elements in schema T
 One-to-one matches
 Movies.title/Books.title = Items.name
 Products.rating/Books.category = Items.classification
 One-to-many matches
 Items.price = Products.basePrice * (1 + Locations.taxRate)
 Other types of matches
 many-to-one, many-to-many

11
Semantic Mappings
 Semantic mapping between Aggregator, Book-vendor, DVD-vendor
 GAV mapping shows how to obtain an entire tuple for Items table of
AGGREGATOR
Create View Items AS
SELECT title AS name, releaseDate AS releaseInfo, rating AS
classification, basePrice * (1 + taxRate) AS price
FROM Movies, Products, Locations
WHERE Movies.id = Products.mid AND Products.saleLocID = Locations.lid
UNION
SELECT title AS name, PubYear AS releaseInfo, category AS classification,
markedPrice AS price
FROM Books, Shop
WHERE Books.ISBN=Shop.BID;
 If we use Local-as-View (LAV) approach to relate schemas
 for each table in DVD-VENDOR and BOOK-VENDOR, must create a semantic mapping that
specifies how to obtain tuples for that table from schema AGGREGATOR (i.e., from table Items)
 If we use GLAV approach 12
 there are semantic mappings going in both directions
Challenges of
Schema Matching and Mapping
 Matching and mapping systems must reconcile semantic
heterogeneity between the schemas
 Such semantic heterogeneity arise in many ways
 same concept, but different names for tables and attributes
 rating vs classification
 multiple attributes in 1 schema relate to 1 attribute in the other
 basePrice and taxRate relate to price
 tabular organization of schemas can be quite different
 one table in AGGREGATOR vs three tables in DVD-VENDOR
 coverage and level of details can also differ significantly
 DVD-VENDOR also models releaseDate and releaseCompany

13
Challenges of
Schema Matching and Mapping
 Why do we have semantic heterogeneity?
 schemas are created by different people whose states and
styles are different
 disparate databases are rarely created for exact same purposes
 Why reconciling semantic heterogeneity is hard
 the semantics is not fully captured in the schemas
 schema clues can be unreliable
 intended semantics can be subjective
 correctly combining the data is difficult

14
Matching System Architecture

15
Matchers

 schemas  similarity matrix

 Input: two schemas S and T, plus any possibly helpful
auxiliary information (e.g., data instances, text
descriptions)
 Output: sim matrix that assigns to each element pair of S
and T a number in [0,1] predicting whether the pair
match
 Numerous matchers have been proposed
 We describe a few, in two classes:
name matchers and data matchers
16
Name-Based Matchers

 Use string matching techniques

 e.g., edit distance, Jaccard, Soundex, etc.
 Often have to pre-process names
 split them using certain delimiters
 e.g., saleLocID  sale, Loc, ID
 expand known abbreviations or acronyms
 loc  location, cust  customer
 expand a string with synonyms / hypernyms
 add cost to price, expand product into book, dvd, cd
 remove stop words
 in, at, and

17
Example

18
Instance-Based Matchers

 When schemas come with data instances, these can be

extremely helpful in deciding matches
 Many instance-based matchers have been proposed
 Some of the most popular
 recognizers
 use dictionaries, regexes, or simple rules
 overlap matchers
 examine the overlap of values among attributes
 Typically applies to attributes whose values are drawn from some finite
domain
 classifiers
 use learning techniques
19
Building Recognizers

 Use dictionaries, regexes, or rules to recognize data

values of certain kinds of attributes
 Example attributes for which recognizers are well suited
 country names, city names, Indian states
 person names (can use dictionaries of last and first names)
 color, rating (e.g., G, PG, PG-13, etc.), phone, PAN
 genes, protein, PIN codes

20
Measuring the Overlap of Values

 Typically applies to attributes whose values are drawn

from some finite domain
 e.g., movie ratings, movie titles, book titles, country names
 Jaccard measure is commonly used
 Example:
 use Jaccard measure to build a
data-based matcher between
DVD-VENDOR and AGGREGATOR
 AGGREGATOR.name refers to DVD titles,
DVD-VENDOR.name refers to sale locations,
DVD-VENDOR.title refers to DVD titles
 low score for (name, name), high score for (name, title)

21
Using Classifiers

 Builds classifiers on one schema and uses them to

classify the elements of the other schema
 e.g., use Naïve Bayes, decision tree, rule learning, SVM
 A common strategy
 for each element si of schema S, want to train classifier Ci to
recognizer instances of si
 to do this, need positive and negative training examples
 take all data instances of si (that are available) to be positive examples
 take all data instances of other elements of S to be negative examples

 train Ci on the positive and negative examples

22
Using Classifiers

 A common strategy (cont.)

 now we can use Ci to compute sim score between si and each
element tj of schema T
 to do this, apply Ci to data instances of tj
 for each instance, Ci produces a number in [0,1] that is the confidence
that the instance is indeed an instance of si
 now need to aggregate the confidence scores of the instances
(of tj) to return a single confidence score (as the sim score
between si and tj)
 a simple way to do so is to compute the average score over all
instances of tj

23
Using Classifiers: An Example

 si is address, tj is location
 Sim scores are 0.9, 0.7, and 0.5, respectively for the three
instances of T.location  return average score of 0.7 as
sim score between address and location
24
Reminder: Matching System Architecture

25
Combining Match Predictions

26
Combining Match Predictions: Another
Example of the Average Combiner

27
Combining Match Predictions

 When to use which combiner?

 average combiner : when we do not have any reason to trust
one matcher over the others
 maximum combiner: when we trust a strong signal from
matchers, i.e., if a matcher outputs a high value, we are
relatively confident that the two elements match
 minimum combiner: when we want to be more conservative
 More complex types of combiners
 use hand-crafted scripts
 e.g., if si is address, return the score of the data-based matcher
otherwise, return the average score of all matchers
 weighted-sum combiners
28
Reminder: Matching System Architecture

29
Enforcing Domain Integrity Constraints

 Designer often has knowledge that can be naturally

expressed as domain integrity constraints
 Constraint enforcer exploits these to prune certain match
combinations
 searches through the space of all match combinations
produced by the combiner
 finds one combination with the highest aggregated confidence
score that satisfies the constraints

30
Enforcing Domain Integrity Constraints -- contd

 Suppose designer knows that

 AGGREGATOR.name refers to movie titles
 many movie titles contain at least four words
 Designer can specify a constraint such as
 if an attribute A matches AGGREGATOR.name, then in any
random sample of 100 data values of A, at least 10 values must
contain four words or more
 Now the constraint enforcer can search for the best match
combination that satisfies this constraint.

31
Schema mapping

32
Overview of Mapping Systems

 Input: matches, output: actual mappings

 Key challenge: find how tuples from one source can be
transformed and combined to produce tuples in the
other
 which data transformation to apply?
 which joins to take?
 and many more possible decisions

33
 Multiple Join Paths
Address id Addr

Professor id name salary Personnel

Student name GPA Yr Sal

PayRate Rank HrRate

WorksOn name Proj hrs ProjRank

f1: PayRate(HrRate) * WorksOn(Hrs) = Personnel(Sal)

 Two Possible Queries
select P.HrRate * W.hrs
from PayRate P, WorksOn W
where P.Rank = W.ProjRank

select P.HrRate * W.hrs

from PayRate P, WorksOn W, Student S
where W.Name=S.Name and
S.Yr = P.Rank

What join path to choose from the candidate sets?

 Horizontal partitioning
Address id Addr

Professor id name salary Personnel

Student name GPA Yr Sal

PayRate Rank HrRate

WorksOn name Proj hrs ProjRank

f2: Professor(Sal)  Personnel(Sal)

 What Kind of Union?
select P.HrRate * W.hrs
from PayRate P, WorksOn W
where P.Rank = W.ProjRank

UNION ALL
select Sal
from Professor
Could also do an outer-union, if Personnel(id,name, sal)…
and even a join.