IR Models & File Structures - 1

• Documents, structure (later units)
• Terms/Concepts/Authority Info.
  • Dictionary, lexicon, thesaurus (U-KB)
  • Access (U-IR): hash table, B-tree
• Document access (later units)
  • Operations and structures: ...

IR Models & File Structures - 2

• Document access structures
  • Graph: hypertext (U-HT)
  • Tree: clustered files (U-CL)
  • String: string search (U-SS))
  • Sequential file
    1. Unordered: signature file
    2. Ordered (Ch 3, 5): ...

IR Models & File Structures - 3

• Document access operations:
  1. Sequential file - Ordered:
     • Build IF, Lookup term (Ch 3)
     • Manipulate term sets (Ch 12)
     • Compute query-doc sim (Ch 15)

Boolean Set Processing / Venn Diagram

Boolean Set Processing / Inverted File Results

• Inverted file is called that since looks at matrix by term instead of by document (the normal case for readers).
• Extension to handle proximity takes extra space.
  • Entries for a term might include (doc,wt) pairs plus:
  • List of locations inside document as:
  • Byte offset (or offsets for start/end); or
  • Paragraph number, sentence number, word number; or
  • Pointer into a structure tree (or pointers for a span):
    • e.g., chapter no. / section no. / subsection no. / par. no.;
    • e.g., reference no. / title part / subtitle field;
    • e.g., dictionary headword / part of speech / sense / definition.

Online Searching - 1

• Phases
  • Clarify info. need / problem
  • Identify access points: a, t, s, ...
  • Identify concepts, terms
  • Develop, try, adapt search strategies
  • Examine results, use feedback

Online Searching - 2

• Query organization
  • String of pearls (DNF)
  • List of required concepts (CNF)
• Concept organization building on:
  • Elements: descriptors, phrases, words, stems/roots
  • Relationships: Synonym, xref, bt

Boolean Operations (Ch 12) - 1

• Set is for a term
• Set element represents a document
• Document representations (keys)
  • Bit vector: doc # (e.g., 1 ... N= 256)
  • Hash table: doc. name or id that is hashed into some bucket
• Iterate is like LISP map functions

Boolean Operations - 2

• Set is for a term
• Set element represents a document
• Document representations (keys)
  • Bit vector: doc # (e.g., 1 ... N= 256)
  • Hash table: doc. name or id that is hashed into some bucket
• Iterate is like LISP map functions

Boolean Operations - 3

• Use of operations for IS&R
  • Lookup term in IF, create set for it, enter elements (docs) in the set
  • Operate (union, ...) on sets
  • Iterate on result set for output
• Table 12.2 terminology
  • Domain size = possible # docs
  • No. elements = actual # docs

Preview of Unit 4

• PAT 3.3 User's Guide
  • copies in library
  • Lectures
    • PAT, string searching
  • Exercises, Lab: PAT, visualization