The Simplest Query Language That Could Possibly Work

1
The Simplest Query Language That Could Possibly
Work

• Richard A. OKeefe
• Andrew Trotman
• Department of Computer Science
• University of Otago
• Dunedin, New Zealand

2
Background
3
Old and New

• Old
• INEX02 and INEX03 use different languages
• Currently
• Extensions to XPath 1.0
• Database language
• Hard and complex
• New
• Proposal for INEX04 and future INEXes
• Designed for INEX
• Information retrieval query language
• Simple, extensible, clearly defined

4
Usage Problems

• 19 CAS topics needed title corrections
• Thats 63
• 3 CO topics needed title corrections
• Thats 8
• Corrections took 12 rounds
• Thats 40 days of corrections
• Proved hard to implement
• 5 institutions submitted correct topics
• topics were submitted with answers
• yes, we submitted bad ones too

5
XPath issues

• Overkill
• Tag instancing never used e.g. p2
• Child / not needed descendant was better //
• Over design
• ltfngtJoelt/fngtltlngtBloggslt/lngt
• Gives string JoeBloggs not Joe Bloggs
• ltyrgt2000,lt/yrgt
• Is not a number
• Operators must be followed faithfully
• AND, OR, lt, gt, etc.
• Could progress in directions not for us

6
DB and IR are different

• Languages based on a database view
• SQL, OQL, etc.
• DB and IR are different

7
So Now What?

• What we need
• Tag equivalence
• Precise syntax for imprecise queries
• Not tied to one approach
• Not tied to DB, IR, Probabilistic, etc
• Extensible for future INEXes

8
Massaging the XML
9
Architectural Forms

• Idea taken from HyTime
• Extend the DTD
• lt!ENTITY old-dtd PUBLIC "..." "oldarticle.dtd"gt
• old-dtd
• lt!ATTLIST ...gt
• ...
• lt!ATTLIST ...gt
• Add three new attributes
• INEXname - what to do with tags
• INEXatts - what to do with attributes
• INEXscan - how to index tags

10

• INEXname
• lt!ATTLIST ip1 INEXname NMTOKEN FIXED "p"gt
• Process all ltip1gt tags as if they were ltpgt tags
• This gives us same power as tag equivalences, but
  definable in the DTD before the documents are
  indexed
• ltpgt restrictions will not return ltlpgt answers

11

• INEXatts
• lt!ATTLIST foo INEXatts NMTOKEN FIXED "yr year"gt
• Process all yr attributes of foo as if they were
  year attributes of foo
• lt!ATTLIST au INEXatts NMTOKEN FIXED "sequence
  -"gt
• Do not index sequence attributes of au tags
• Tag equivalence now apply to attributes too
• Some attributes are not indexed (so cant be
  found as answers)

12

• INEXscan
• lt!ATTLIST p INEXscan NMTOKEN FIXED "all"gt
• Index the contents of a p tag as belonging to
  that p tag
• Index as expected
• lt!ATTLIST scp INEXscan NMTOKEN FIXED "content"gt
• Index the content of scp tags as if the tag were
  not present
• Removes the ltstgtVltscpgtoicelt/scpgtlt/stgt problem
• lt!ATTLIST ref INEXscan NMTOKEN FIXED "nothing"gt
• Do not index this tag or the contents of this
  tag
• Removes the ltrefgt3lt/refgt problem
• lt!ATTLIST art INEXscan NMTOKEN FIXED "element"gt
• Index this tag, not its content, but do so its
  children
• Included as the remaining case

13
Searching the XML
14
Query Syntax

• The complete grammar is given in the paper
• A (electron)
• answer about A
• PA sec(electron)
• return P tags about A
• PAQB art(density)sec(electron)
• Where P is about A, return P/Q tags about B
• And so on

15
P and Q syntax

• Tags are / (descendant) separated
• art/sec
• means sec descendant of art tag
• There is no child, its not needed
• Attributes cant have descendants so
• art/au_at_sequence
• means the sequence attribute of an au tag
  anywhere under an art tag

16
P and Q Filtering

• artyr 1990..2000
• Restrict art to where the yr tag is between 1990
  and 2000 (whatever the IR system chooses that to
  mean)
• Also
• artyr 1990
• artyr ..2000
• artyr 1990,1993..
• .. Syntax used as it is NOT strict arithmetic,
  its what the IR system chooses it means

17
A and B syntax

• Search terms can be mandatory (), exclusionary
  (-) or optional
• fox in -sox
• Means must include in, must not include sox
  and can include fox. Again, these are hints to
  the IR system.
• Phrases are quoted (single or double)
• knox in box
• knox in box

18
Operators

• These can then be joined with operators
• a b
• a or b or both should hold true
• a b
• a and b should both hold true
• a
• a should not hold true
• ab
• a should appear in b
• Operators are also not precise (hence should
  hold true)

19
Examples

• Topic 61
• //articleabout(.,'clustering distributed') and
  about(.//sec,'java')
• article(clustering distributed javasec)
• Topic 64
• //articleabout(./, 'hollerith')//secabout(./,
  'DEHOMAG')
• article(hollerith)sec(DEHOMAG)
• Topic 66
• /article./fm//yr lt '2000'//secabout(.,'"sear
  ch engines"')
• articlefm/yr ..1999sec("search engines")

20
Examples

• Topic 76
• //article(./fm//yr'2000' OR ./fm//yr'1999')
  AND about(., '"intelligent transportation
  system"')secabout(., 'automation vehicle')
• articlefm/yr 1999..2000("intelligent
  transportation system")sec(automation vehicle)
• Topic 91
• Internet traffic
• (Internet traffic)

21
Questions?