Semistructured Data Seach

Semistructured
Data Search
Krisztian Balog
University of Stavanger

Promise Winter School 2013 | Bressanone, Italy, February 2013

The data landscape

Flat text HTML XML RDF Databases

unstructured data semistructured data structured data

- Semistructured data
- Lack of ﬁxed, rigid schema
- No separation between the data and the schema,
self-describing structure (tags or other markers)

Motivation
- Supporting users who cannot express their
need in structured query languages
- SQL, SPARQL, Inquery, etc.
- Dealing with heterogeneity
- Users are unaware of the schema of the data
- No single schema to the data

Semistructured data
- Advantages
- The data is not constrained by a ﬁxed schema
- Flexible (the schema can easily be changed)
- Portable
- Possible to view structured data as semistructured
- Disadvantages
- Queries are less efﬁcient than in a constrained
structure

In this talk
- How to exploit the structure available in the
data for retrieval purposes?
- Diﬀerent types of structure
- Document, query, context
- Working in a Language Modeling setting
- Number of diﬀerent tasks
- Retrieving entire documents
- I.e., no element-level retrieval
- Textual document representation is readily available
- No aggregation over multiple documents/sources

Incorporating structure

3
context

1 document query 2

Preliminaries
Language modeling

Language Modeling
- Rank documents d according to their likelihood
of being relevant given a query q: P(d|q)
P (q|d)P (d)
P (d|q) = / P (q|d)P (d)
P (q)

Query likelihood Document prior
Probability that query q Probability of the document
was “produced” by document d being relevant to any query

Y
P (q|d) = P (t|✓d ) n(t,q)

t2q

Language Modeling
Query likelihood scoring
Number of times t appears in q

Y
P (q|d) = P (t|✓d ) n(t,q)

t2q
Document language model
Multinomial probability distribution Smoothing parameter
over the vocabulary of terms

P (t|✓d ) = (1 )P (t|d) + P (t|C)
Empirical Maximum Collection
likelihood
document model estimates
model
P
n(t, d) d n(t, d)
P
|d| d |d|

Language Modeling
Estimate a multinomial Smooth the distribution
probability distribution with one estimated from
from the text the entire collection

P (t|✓d ) = (1 )P (t|d) + P (t|C)

Example
In the town where I was born,
Lived a man who sailed to sea,
And he told us of his life,
In the land of submarines,
So we sailed on to the sun,
Till we found the sea green,
And we lived beneath the waves,
In our yellow submarine,
We all live in yellow submarine,
yellow submarine, yellow
submarine, We all live in yellow
submarine, yellow submarine,
yellow submarine.

who
where
waves
us
town
Empirical document LM

told
till
sun
submarines
so
our
man
n(t, d)

life
|d|

land
i
his
P (t|d) =

he
green
found
born
beneath
sea
sailed
lived
live
all
we
yellow
submarine

0.09

0.06

0.03

0
0.15

0.12

Scoring a query
q = {sea, submarine}

P (q|d) = P (“sea”|✓d ) · P (“submarine”|✓d )

Scoring a query
0.03602

0.9 0.04 0.1 0.0002
(1 )P (“sea”|d) + P (“sea”|C)

t P(t|d) t P(t|C)
submarine 0.14 submarine 0.0001
sea 0.04 sea 0.0002
... ...

Scoring a query
0.04538 0.03602 0.12601

0.9 0.14 0.1 0.0001
(1 )P (“submarine”|d) + P (“submarine”|C)

t P(t|d) t P(t|C)
submarine 0.14 submarine 0.0001
sea 0.04 sea 0.0002
... ...

In this part
- Incorporate document structure into the
document language model
- Represented as document ﬁelds

Y
P (d|q) / P (q|d)P (d) = P (d) P (t|✓d )n(t,q)
t2q

Document
language model

Use case
Web document retrieval

Unstructured representation
PROMISE Winter School 2013
Bridging between Information Retrieval and Databases
Bressanone, Italy 4 - 8 February 2013
The aim of the PROMISE Winter School 2013 on "Bridging between Information
Retrieval and Databases" is to give participants a grounding in the core
topics that constitute the multidisciplinary area of information access and
retrieval to unstructured, semistructured, and structured information. The
school is a week-long event consisting of guest lectures from invited
speakers who are recognized experts in the field. The school is intended for
PhD students, Masters students or senior researchers such as post-doctoral
researchers form the fields of databases, information retrieval, and related
fields.
[...]

HTML source
<html>
<head>
<title>Winter School 2013</title>
<meta name="keywords" content="PROMISE, school, PhD, IR, DB, [...]" />
<meta name="description" content="PROMISE Winter School 2013, [...]" />
</head>
<body>
<h1>PROMISE Winter School 2013</h1>
<h2>Bridging between Information Retrieval and Databases</h2>
<h3>Bressanone, Italy 4 - 8 February 2013</h3>
<p>The aim of the PROMISE Winter School 2013 on "Bridging between
Information Retrieval and Databases" is to give participants a grounding
in the core topics that constitute the multidisciplinary area of
information access and retrieval to unstructured, semistructured, and
structured information. The school is a week-long event consisting of
guest lectures from invited speakers who are recognized experts in the
field. The school is intended for PhD students, Masters students or
senior researchers such as post-doctoral researchers form the fields of
databases, information retrieval, and related fields. </p>
[...]
</body>
</html>

Fielded representation based on HTML markup
title: Winter School 2013

meta: PROMISE, school, PhD, IR, DB, [...]
PROMISE Winter School 2013, [...]

headings: PROMISE Winter School 2013
Bridging between Information Retrieval and Databases
Bressanone, Italy 4 - 8 February 2013

body: The aim of the PROMISE Winter School 2013 on "Bridging between
Information Retrieval and Databases" is to give participants a
grounding in the core topics that constitute the multidisciplinary
area of information access and retrieval to unstructured,
semistructured, and structured information. The school is a week-
long event consisting of guest lectures from invited speakers who
are recognized experts in the field. The school is intended for
PhD students, Masters students or senior researchers such as post-
doctoral researchers form the fields of databases, information
retrieval, and related fields.

Fielded Language Models
[Ogilvie & Callan, SIGIR’03]

- Build a separate language model for each ﬁeld
- Take a linear combination of them
m
X
P (t|✓d ) = µj P (t|✓dj )
j=1

Field language model
Smoothed with a collection model built
from all document representations of the
same type in the collection

Field weights
m
X
µj = 1
j=1

Field Language Model
Smoothing parameter

P (t|✓dj ) = (1 j )P (t|dj ) + j P (t|Cj )

Empirical Maximum Collection
likelihood
ﬁeld model estimates
ﬁeld model
P
n(t, dj ) d n(t, dj )
P
|dj | d |dj |

Fielded Language Models
Parameter estimation

- Smoothing parameter
- Dirichlet smoothing with avg. representation length
- Field weights
- Heuristically (e.g., proportional to the length of text
content in that ﬁeld)
- Empirically (using training queries)
- Computationally intractable for more than a few ﬁelds

Example
q = {IR, winter, school}
ﬁelds = {title, meta, headings, body}
µ = {0.2, 0.1, 0.2, 0.5}

P (q|✓d ) = P (“IR”|✓d ) · P (“winter”|✓d ) · P (“school”|✓d )

P (“IR”|✓d ) = 0.2 · P (“IR”|✓dtitle )
+ 0.1 · P (“IR”|✓dmeta )
+ 0.2 · P (“IR”|✓dheadings )
+ 0.2 · P (“IR”|✓dbody )

Use case
Entity retrieval in RDF data

Use case
Entity retrieval in RDF data
dbpedia:Audi_A4

foaf:name Audi A4
rdfs:label Audi A4
rdfs:comment The Audi A4 is a compact executive car
produced since late 1994 by the German car
manufacturer Audi, a subsidiary of the
Volkswagen Group. The A4 has been built [...]
dbpprop:production 1994
2001
2005
2008
rdf:type dbpedia-owl:MeanOfTransportation
dbpedia-owl:Automobile
dbpedia-owl:manufacturer dbpedia:Audi
dbpedia-owl:class dbpedia:Compact_executive_car
owl:sameAs freebase:Audi A4
is dbpedia-owl:predecessor of dbpedia:Audi_A5
is dbpprop:similar of dbpedia:Cadillac_BLS

Hierarchical Entity Model
[Neumayer et al., ECIR’12]

- Number of possible fields is huge
- It is not possible to optimise their weights directly
- Entities are sparse w.r.t. different fields
- Most entities have only a handful of predicates
- Organise fields into a 2-level hierarchy
- Field types (4) on the top level
- Individual fields of that type on the bottom level
- Estimate field weights
- Using training data for field types
- Using heuristics for bottom-level types

Two-level hierarchy
foaf:name Audi A4
Name rdfs:label Audi A4
rdfs:comment The Audi A4 is a compact executive car
produced since late 1994 by the German car
manufacturer Audi, a subsidiary of the
Volkswagen Group. The A4 has been built [...]
Attributes dbpprop:production 1994
2001
2005
2008
rdf:type dbpedia-owl:MeanOfTransportation
dbpedia-owl:Automobile
Out-relations dbpedia-owl:manufacturer dbpedia:Audi
dbpedia-owl:class dbpedia:Compact_executive_car
owl:sameAs freebase:Audi A4
is dbpedia-owl:predecessor of dbpedia:Audi_A5
In-relations is dbpprop:similar of dbpedia:Cadillac_BLS

Field generation
- Uniform
- All fields of the same type are equally important
- Length
- Proportional to field length (on the entity level)
- Average length
- Proportional to field length (on the collection level)
- Popularity
- Number of documents that have the given field

Comparison of models

t df t F df t
d ... d ... ... d ... ... ...

t df t F df t

Unstructured Fielded Hierarchical
document model document model document model

Use case
Finding movies in IMDB data

Use case
Finding movies in IMDB data

<title>The Transporter</title>
<year>2002</year>
<language>English</language>
<genre>Action</genre>
<genre>Crime</genre>
<genre>Thriller</genre>
<country>USA</country>
<actors>
<actor>Jason Statham</actor>
<actor>Matt Schulze</actor>
<actor>François Berléand</actor>
<actor>Ric Young</actor>
<actress>Qi Shu/actress>
</actors>
<team>
<director>Louis Leterrier</director>
<director>Corey Yuen</director>
<writer>Luc Besson</writer>
<writer>Robert Mark Kamen</writer>
<producer>Luc Besson</producer>
<cinematographer>Pierre Morel</cinematographer>
</team>

Probabilistic Retrieval Model
for Semistructured data
[Kim et al., ECIR’09]
- Find which document ﬁeld each query term
may be associated with
- Extending [Ogilvie & Callan, SIGIR’03]
m
X
P (t|✓d ) = µj P (t|✓dj )
j=1
Mapping probability
Estimated for each query term
m
X
P (t|✓d ) = P (dj |t)P (t|✓dj )
j=1

PRMS
Mapping probability
P
d n(t, dj )
P (t|Cj ) = P
d |dj |

Prior field probability
Term likelihood Probability of mapping the query term
Probability of a query term to this field before observing collection
occurring in a given field type statistics

P (t|dj )P (dj )
P (dj |t) =
P (t)

X
P (t|dk )P (dk )
dk

PRMS
Mapping example

Query: meg ryan war

dj P (t|dj ) dj P (t|dj ) dj P (t|dj )
cast 0.407 cast 0.601 genre 0.927
team 0.382 team 0.381 title 0.070
title 0.187 title 0.017 location 0.002

Structured query
representations
- Query may have a semistructured
representation, i.e., multiple ﬁelds
- Examples
- TREC Genomics track
- (1) gene name, (2) set of symbols
- TREC Enterprise track, document search task
- (1) keyword query, (2) example documents
- INEX Entity track
- (1) keyword query, (2) target categories, (3) example entities
- TREC Entity track
- (1) keyword query, (2) input entity, (3) target type

Use case
Enterprise document search (TREC 2007)

- Task: create an overview page on a given topic
- Find documents that discuss the topic in detail

cancer risk +
<topic>
<num>CE-012</num>
<query>cancer risk</query>
<narr>
Focus on genome damage and therefore cancer risk in humans.
</narr>
<page>CSIRO145-10349105</page>
</topic>

Retrieval model
- Maximizing the query log-likelihood provides
the same ranking as minimising KL-divergence
- Assuming uniform document priors

P (t|✓d ) KL(✓q ||✓d ) P (t|✓q )

Document Query
model model

Estimating the query model
- Baseline maximum-likelihood
n(t, q)
P (t|✓q ) = P (t|q) =
|q|
- Query expansion using relevance models
[Lavrenko & Croft, SIGIR’01]
P (t|✓q ) = (1 )P (t|ˆ) + P (t|q)
q
Expanded query model
Based on term co-occurrence statistics

P (t, q1 , . . . , qk )
P (t|ˆ) ⇡ P
q
0 P (t , q1 , . . . , qk )
0
t

Sampling from examples
[Balog et al., SIGIR’08]

d
P (t|d) P (t|S) P (t|ˆ)
q
... P (d|S) top K
d
d terms

sampling expanded
sample distribution query
documents

X
P (t|S)
P (t|S) = P (t|d)P (d|S) P (t|ˆ) = P
q
0 2K P (t |S)
0
t
d2S

Term importance Document importance

Importance of a sample document

- Uniform P (d|S) = 1/|S|
- All sample document are equally important
- Query-biased P (d|S) / P (d|q)
- Proportional to the document’s relevance to the
(original) query
- Inverse query-biased P (d|S) / 1 P (d|q)
- Reward documents that bring in new aspects (not
covered by the original query)

Estimating term importance

- Maximum-likelihood estimate
n(t, d)
P (t|d) =
|d|
- Smoothed estimate
P (t|d) = P (t|✓d ) = (1 )P (t|d) + P (t|C)

- Ranking function by [Ponte, 2000]
X P (t|d)
P (t|d) = s(t)/ s(t )
0
s(t) = log
t0
P (t|C)

Use case
Entity retrieval in Wikipedia (INEX 2007-09)

- Given a query, return a ranked list of entities
- Entities are represented by their Wikipedia page
- Entity search
- Topic deﬁnition includes target categories
- List search
- Topic deﬁnition includes example entities

Example query

<title>Movies with eight or more Academy Awards</title>
<categories>
<category id="45168">best picture oscar</category>
<category id="14316">british films</category>
<category id="2534">american films</category>
</categories>

Using categories for retrieval

- As a separate document ﬁeld
- Filtering
- Similarity between target and entity categories
- Set similarity metrics
- Content-based (concatenating category contents)
- Lexical similarity of category names

Also related to categories
- Category expansion
- Based on category structure
- Using lexical similarity of category names
- Ontology-base expansion
- Generalisation
- Automatic category assignment

Modeling terms and categories
[Balog et al., TOIS’11]

P (e|q) / P (q|e)P (e)
P (q|e) = (1 )P (✓q |✓e ) + P (✓q |✓e )
T T C C

Term-based representation Category-based representation
Query model Entity model Query model Entity model

p(t|✓q )
T
p(t|✓e )
T
p(c|✓q )
C
p(c|✓e )
C
KL(✓q ||✓e )
T T
KL(✓q ||✓e )
C C

Other types of structure
- Link structure
- Linguistic structure
- Social structures

Link structure

Source: Wikipedia

Link structure
- Aim
- High number of inlinks from pages relevant to the
topic and not many incoming links from other pages
- Retrieve an initial set of documents, then rerank

P (d|q) / P (q|d)P (d)

Document prior
Probability of the document
being relevant to any query

Document link degree
[Kamps & Koolen, ECIR’08]

Local indegree
Number of incoming links from within the
top ranked documents retrieved for q

Ilocal (d)
P (d) / 1 +
1 + Iglobal (d)

Global indegree
Number of incoming links
from the entire collection

Relevance propagation
[Tsikrika et al., INEX’06]

- Model a user (random surfer) that after seeing
the initial set of results
- Selects one document and reads its description
- Follows links connecting entities and reads the
descriptions of related entities
- Repeats it N times

P0 (d) = P (q|d)
X
Pi (d) = P (q|d)Pi 1 (d) + (1 P (q|d0 ))P (d0 |d)Pi 1 (d0 )
d0 !d

The probability of staying at Transition probabilities
the node equals to its Outgoing links from d’ to d set to uniform
relevance to the query

Relevance propagation
[Tsikrika et al., INEX’06]

- Weighted sum of probabilities at diﬀerent steps
N
X
P (d) / µ0 P0 (d) + (1 µ0 ) µi Pi (d)
i=1

Linguistic structure

Source: wisuwords.com

Translation Model
[Berger & Lafferty, SIGIR’99]

Y X n(t,q)
P (q|d) = P (t|w)P (w|✓d )
t2q w2V

Translation model
Probability that word w can
“semantically translated” to word qi

- Obtaining the translation model
- Exploiting WordNet word co-occurrences [Cao et al.,
SIGIR’05]

Use case
Expert ﬁnding

- Given a keyword query, return a ranked list of
people who are experts on the given topic
- Content-based methods can return the most
knowledgeable persons
- However, when it comes to contacting an
expert, social and physical proximity matters

Expert proﬁle
<person>
<anr>710326</anr>
<name>Toine M. Bogers</name>
<name>Toine Bogers</name>
<name>A. M. Bogers</name>
<job>PhD student</job>
<faculty>Faculty of Humanities</faculty>
<department>Department of Communication and Information Sciences</department>
<room>Room D 348</room>
<address>P.O. Box 90153, NL-5000 LE Tilburg, The Netherlands</address>
<tel>+31 13 466 245</tel>
<fax>+31 13 466 289</fax>
<homepage>http://ilk.uvt.nl/~toine</homepage>
<email>A.M.Bogers@uvt.nl</email>

<publications>
<publication arno-id="1212">
<title>Design and Implementation of a University-wide
Expert Search Engine</title>
<author>R. Liebregts and T. Bogers</author>
<year>2009</year>
<booktitle>Proceedings of the 31st European Conference on Information
Retrieval</booktitle>
</publication>
[...]
</publications>
[...]
</person>

User-oriented model for EF
[Smirnova & Balog, ECIR’11]

S(e|u, q) = (1 )K(e|u, q) + T (e|u)

Knowledge gain Contact time
Diﬀerence between the Distance between the user and
knowledge of the expert and that the expert in a social graph
of the user on the query topic

Social structures
Organisational hierarchy

<person>
<anr>710326</anr>
<name>Toine M. Bogers</name> University
[...]
<faculty>Faculty of Humanities</faculty>
<department>Department of Communication Faculty
and Information Sciences</department>
[...]
</person>
Department

Toine M. Bogers
710326

Social structures
Geographical information

<person>
<anr>710326</anr>
<name>Toine M. Bogers</name> Campus
[...]
<room>Room D 348</room>
[...] Building
</person>

Floor

Toine M. Bogers
710326

Social structures
Co-authorship

<person>
<anr>710326</anr>
<name>Toine M. Bogers</name>
[...]
<publications>
<publication arno-id="1212">
<title>
Design and Implementation of a
University-wide Expert Search Engine Toine M. Bogers
</title> 710326
<author>
R. Liebregts and T. Bogers
</author>
<year>2009</year>
<booktitle>
Proceedings of the 31st European
Conference on Information Retrieval
</booktitle>
</publication>
[...]
</publications>
[...]
</person>

Social structures
Combinations
University

Department Department

Faculty Faculty Faculty

Toine M. Bogers
710326

Floor Floor

Building Building

Campus

Summary
- Diﬀerent types of structure
- Document structure
- Query structure
- Contextual structures
- Probabilistic IR and statistical Language
Models yield a principled framework for
representing and exploiting these structures

Questions?

Contact | @krisztianbalog | krisztianbalog.com

Semistructured Data Seach

Recommended

More Related Content

More from krisztianbalog (19)

Recently uploaded (20)

Semistructured Data Seach