Dias nummer 1

1
Bibliographic relationships
Erik Thorlund Jepsen The Danish Library Agency
2
Outline

• Bibliographic relations
• Definitions and types
• Importance of bibliographic relationships
• Searching (navigating), identifying and
  selecting
• Typologies
• Utilizing bibliographic relationships in OPACs
  and other search tools (eg. Integrated search)
• By cataloguing
• By automatic means
• Emploing statistical measures
• Conclusion

3
Definition of Bibliographic relationships

• A relationship between information entities
• exist, when two entities are somehow
• associated with each other.
• (Velucci, 1997, p.105).
• Though semantically precise, this definition does
  not provide much guidance for the identification
  of relationships, since associations rely on
  subjective judgements, assessing the relevance of
  connecting/relating two or more entities.

4
Bibliographic entities (FRBR)
5
Relations and associations

• Who identifies a relation (associates)
• Authors, publishers, indexers/cataloguers, users,
  system/rule based (incl. user tracks)
• What are the associations based on?
• Information in entities?
• About another entity
• About a relation between two other entities
• Similarity
• Use (and users)

6
Examples

• Book is based on ..... (work, expression,
  manifestation)
• Book is 3. Edition of ....
• In his book, Peter Ingwersen mentions the
  findings of Pia Borlund in .... (work,
  expression, manifestation)
• A novel is part of an anthology
• Book X and Y are written by the same author
• Book X and Y shares two descriptors
• Person A lends book X and CD Y
• Article X and article Y is cited in article
  Z
• Book X and Y share a lot of words
• Book X and Y share a lot of references

7
Importance of relations

• A little bit about FRBR and relations
• Examples
• Identification and selection
• Use of linkages
• Faceted search facilities
• Importance as stated by FRBR and GFOD

8
FRBR - tasks

• to find entities that correspond to the user's
  stated search criteria (i.e., to locate either a
  single entity or a set of entities in a file or
  database as the result of a search using an
  attribute or relationship of the entity)
• to identify an entity (i.e., to confirm that the
  entity described corresponds to the entity
  sought, or to distinguish between two or more
  entities with similar characteristics)
• to select an entity that is appropriate to the
  user's needs (i.e., to choose an entity that
  meets the user's requirements with respect to
  content, physical format, etc., or to reject an
  entity as being inappropriate to the user's
  needs)
• to acquire or obtain access to the entity
  described (i.e., to acquire an entity through
  purchase, loan, etc., or to access an entity
  electronically through an online connection to a
  remote computer).
• (Functional Requirements for Bibliographic
  Records, 1998, p.82)

9
FRBR additional tasks?

• to relate................. A fifth task?
• Even more, FRBR reminds us of the importance of
  bibliographic relationships, and reminds us that
  we describe things in the bibliographic universe
  in order to meet specific user tasks find,
  identify, select, obtain, and i add
  relate (Tillett, 2005, p. 198).
• Yet, information about relationships supports the
  three tasks to find, to identify and to select
  (e.g. supports collocation, which is seen as part
  of to find)
• In other words, to relate is a sub task of to
  find, to identify and to select.
• It could cause a breakdown of the model to
  incorporate to relate as a fifth task
• To navigate.A fifth task?
• Yes often among groups of entities

10
Importance of relations FRBR-terms

• Information about bibliographic relations between
  two or more bibliographic entities can support
  the user tasks find, identify and select.
  Relations stated somehow in a bibliographic
  record can potentially
• Improve users understanding of a given entity,
  which potentially strengthens the identification
  and selection/deselection of the entity.
• Improve users options for finding relevant
  entities by leading the way from a known (found)
  entity too related entities which are more
  relevant in a given situation.

11
Importance of relations

• Furthermore, information about bibliographic
  relationships can strengthen the users
  understanding of the system (database) at hand
  and the knowledge organization in the system, by
• Creating groups of entities
• and
• Facilitating navigation in the bibliographic
  universe (the database/Catalogue)

12
Example identification and selection

• Draznin, Sandra LeighBørnenes restaurant
• Bogen er baseret på tv-serien Børnenes
  restaurant og indeholder enkle madopskrifter som
  8-12 årige selv kan lave. Det gælder både
  forretter, hovedretter, desserter og snacks.
• BOG 1. udgave, 1. oplag. TV 2, 2007
• Opskrifter af Sandra Leigh Draznin, fotos Jes
  Buusmann, opskrifter på side 20, 32, 42, 56 og
  74 Thomas Castberg Larsen, forord og tips ved
  Steffen Bjergved og Thomas Castberg Larsen,
  efterskrift af Carina Christensen1. udgave.
  2007. 92 sider, illustreret i farverForlag TV
  2Form kogebøger opskrifterOpstilling i
  folkebiblioteker 64.1Biblioteket anbefaler Fra
  10 årISBN-13 978-87-92121-13-4Pris ved
  udgivelsen kr. 199,00

13
Example linking
14
(No Transcript)
15
(No Transcript)
16
Example Faceted search
17
(No Transcript)
18
Importance of linking Danish example

• Analysis of searches in bibliotek.dk
• (20.506 searches - 20. December 2004)
• Free-text 7
• Author 34
• Link author 5
• Title 20
• Descriptor/keyword 11
• Link descriptor/keyword more like this
• and Literature about..) 15
• Other (each max 2) sum
  8
• Source Kirsten Larsen, Deputy Head, The Danish
  Library Centre (DBC)

19
Importance GFOD

• User Principle General guidelines for good
  practice in display design and criteria for
  effective screen displays as these relate to
  legibility, clarity, understandability and
  navigability
• Content and Arrangement Principle 7. Support
  navigation from the displayed information to
  related information
• (this principle is further divided into more
  specific, and i add ambitious, principles.)

20
Bibliographic relationships identification of,
and typologies

• Associations/relations can be identified by
  analyzing
• Sets of documents
• Existing information systems
• Standards, rule sets and registration formats
• Empirical studies of users identification - and
  assessment of importance of associations among
  groups of entities

21
Typologies

• Categories Holds between
• Equivalence Relationships (copies, facsimiles,
  microforms and other similar reproductions)
• Derivative Relationships (versions, editions,
  revisions, translations)
• Descriptive Relationships (annotated editions,
  commentaries, reviews)
• Whole-Part Relationships (selections from
  anthologies, collections, series, chapters vs.
  books)
• Accompanying Relationships (supplements,
  concordances, indexes)
• Sequential Relationships (sequels of a monograph,
  parts of a series)
• Shared Characteristics Relationships (common
  author, publisher, title, subject)
• Vellucci and Tillett Categories of Relations
  (Shortened description from
• Vellucci, 1997)

22
Content relationships (equivalence, derivative
and descriptive) are sometimes hard to
distinguish in practice.
23


24
(No Transcript)
25
(No Transcript)
26
Typologi - FRBR

• Relationships depicted in the high level diagrams
• Other Relationships between Group 1 entities at
  these levels
• Work-to-work
• Expression-to-expression
• Expression-to-work
• Manifestation-to-manifestation
• Manifestation-to-item
• Item-to-item
• Whole/Part at work, expression, manifestation and
  item Level
• Not meant to be exhaustive!
• Yet, relationships are mapped to user task
  (alongside attributes)

27
Utilization

• Three purposes when cataloguing information about
  relations and setting up system rules
• Identification and understanding of relation
• Linking from found entity to related entities
• Displaying meaningful/useful sets of records

28
Relations expressed as links

• Relations are expressed as implicit or explicit
  links, where explicit links are divided into
  directional and mechanical links (hyperlinks)
• (Velucci, 1997)
• Hyperlinks are constructed by manual or
  computational means
• Manual links are static and are commonly used to
  structure texts or to connect associatively
  related entities (by topic) (and to connect
  bibliographic families added by etj)
• Computational links can be created at search time
  (dynamicality) and are primarily used to connect
  similar entities (e.g. based on shared
  characteristics added by etj)
• (Agosti, 1997)

29
Traditional means vs. new ways

• Traditionally relations are handled in very
  different ways, caused by
• Different types of relationships are handled
  differently
• Non-specific rules
• Variances between library systems
• Differences in cataloguing policies
• .......

30
Traditional means vs. new ways (2)

• Traditional methods include (very generalized)
• Work, expression, manifestation and (sometimes)
  item information included in one record.
• including notes on predecessors
• often goes for part-whole relations
• Hyperlink structures based on controlled
  information about author, subject and (sometimes
  title) (most commonly links from author(s),
  descriptors and classification numbers)

31
Traditional means vs. new ways (3)

• Initiatives to strengthen the utilization of
  bibliographic relations could be divided into
• Initiatives that try to put the display of
  relationships on the agenda (e.g. GFOD)
• Initiatives that rely on manual work and further
  development of rule sets and registration formats
  (e.g. Reuse)
• Initiatives that add links to the display of
  bibliographic records by controlling data and
  implementing local rules for displays. (ad hoc,
  local or system-specific initiatives)
• Initiatives that try to collocate the different
  expressions and manifestations of the work (e.g.
  Bibliotek.dk)
• Initiatives that try to structure the catalogue
  (existing data) according to FRBR (e.g. FRBR
  Display Tool)
• Faceted search facilities
• The list is not exhaustive!

automa-tic or semi-automa-tic
32
Utilization and links examples

• One record structures (e.g. for accompanying
  relations)
• Computational links for shared characteristics
  (The KB example renæssance)
• Rules and codes (e.g. for derived relations)
• Computational solutions for work display

33
One-record structures
34
Rules and codes example Reuse

• Widened use of specific field in Marc-formats to
  handle relations in a uniform way.
• 787 Non-specific relationship entry
  (Repeatable)....and two subfields
• w Record control number (target to link current
  record to)
• g Relationship information (textual optional)

35
Reuse (2)

• To distinguish between the various relationships,
  and to make them specific, our simple model
  proposes the use of indicator 2 in 787, as yet
  undefined. This indicator might take on the
  following values (and here, a full-scale model
  would not have to differ) (in parentheses DC
  Simple terms for relations)
• 0 Equivalence (facsimile or reproduction)
  (IsFormatOf) 1 Simultaneous edition
  (IsVersionOf) 2 Successive derivation, edition,
  version (IsVersionOf) 3 Amplification (incl.
  commentaries, illustrations, criticism etc.)
  (IsBasedOn) 4 Extraction (abridgements,
  condensations, excerpts) 5 Recordings of
  performances 6 Adaptation, modification (change
  of genre or medium, arrangement) (IsFormatOf) 9
  Translations (IsVersionOf)
• a Accompanying relationship (supplements of any
  kind) (IsRequiredBy) p Part à whole relationship
  (IsPartOf) r Review or other descriptive
  relationship s Sequential relationship (like
  successive title of a serial) u Unspecific
  relationship, based on shared characteristics of
  other kinds
• (Eversberg, 1998)

36
On collocating the work

• Most users seek particular works, not particular
  editions. Yet works are published in the form of
  editions the fundamental duty of descriptive
  cataloguing is to organize the resulting chaotic
  bibliographic universe to facilitate user access
  to works, and to allow them easily to select the
  edition of the work sought that best meets their
  needs (Yee, 1997, p.64).

37
Computational solutions for work display

• FRBR Display Tool
• Library of Congress FRBR Display Tool was
  developed to transform bibliographic data found
  in MARC 21 record files into meaningful displays
  by grouping them into the work, expression and
  manifestation FRBR entities. Based on XML
  technologies, the tool may be altered to meet the
  needs of individual institutions. It also shows
  how the theoretical portion of the FRBR model can
  be used practically to allow librarians to
  evaluate the consistency of their local
  bibliographic data

38
Work-display Bibliotek.dk (The Danish Union
Catalogue)

• An example of an almost totally automatic
  initiative is the display of editions of a work
  in the Danish Union Catalogue Bibliotek.dk
• Attributes like author and title are used in a
  best match algorithm to identify different
  editions of the work.
• Due to, a high level of authority control and the
  use of original titles, the different expressions
  of a work will normally be collocated in the
  search result.

39
bibliotek.dk - library.dk

• End user version of the Danish Union Catalogue
• Sponsored by The library Agency but maintenance
  and development by The Danish Library Centre
  (DBC)
• Content
• The Danish national bibliography
• all titles in public libraries and research
  libraries in Denmark
• Content is not 100 equivalent to the Union
  Catalogue (availability matters)
• Works together with a national transportation
  system users can pick up books from every
  library at their own (chosen) library

40
Adaptation of FRBR in bibliotek.dk

• The records in bibliotek.dk represents
  manifestations (AACR2/danMARC2).
• The aim is to present these records grouped
  according to the work they embody
• At one point our definition differs from FRBR
• For practical reasons we consider expressions in
  different language to be different works.
• You could also say that in this case we prefer
  grouping according to the expression of the work.
• (Paul B. Jensen, Danish Library Center)

41
Implementing the work concept

• The work level display is based on matching and
  collocating manifestation records on-the fly
• This match is based on simple author and title
  data in normalized form
• From the work level you can expand to the
  manifestations, select one (or more) and make a
  request

(Paul B. Jensen, Danish Library Center)
42
Accomplishment

• A more user-friendly interface (as confirmed by a
  majority of test-users)
• A reduction of unnecessary inter-library loans,
  because it is easier to locate an edition to your
  local library (or libraries)
• (Paul B. Jensen, Danish Library Center)

43
Challenges(read problems)

• In principle a traditional aacr2/marc-record does
  not specify which bibliographic information
  refers to work level and which to the
  expression/manifestation level
• Many bibliographic items contains more than one
  work
• Collected plays in one volume (e.g. Shakespeare)
• 3 novels in one volume
• 3 symphonies on one cd
• Etc.
• (Paul B. Jensen, Danish Library Center)

44
(No Transcript)
45
(No Transcript)
46
Neglect or choose edition
47
Show editions
Show full record
48
Other kinds of linkages

• Author-pointers
• citations, references and links
• semantic equivalence (same as similarity below)
• Use-determined
• frequencies
• Similarity-based
• Co-occurrence of text-elements
• e.g. words in text, citations (bibliographic
  coupling)
• Third part pointers
• co-citations
• articles, books ..

49
Types Author pointers
Author
Entities citing, linking to or referring to other
entities
Entity
Entity
Entities cited by, referred by or linked to by
other entities
50
Types Use determined relations
Entity
User
Entities bought or lent by same user
Entity
51
Example use determined links

• RomanSuzanne BrøggerLinda Evangelista Olsen /
  Suzanne Brøgger4. oplag. - Kbh. Gyldendal,
  2002. - 134 siderKatten Linda formodes at være
  en reinkarnation af forfatterens mor, der selv
  var en kat. Og det passer godt nok på den
  tilværelse mor og kat har, og deres måde at
  påvirke omgivelserne på ....Tidligere 1.
  udgave. 2001.Originaludgave 2001.ISBN
  87-00-48736-8 hf. kr. 175,00.
•  Andre, der har lånt Suzanne Brøgger Linda
  Evangelista Olsen, har også låntSuzanne
  Brøgger JaSuzanne Brøgger En gris som har
  været oppe at slås kan man ikke stegeSuzanne
  Brøgger Creme fraicheSuzanne Brøgger
  JadekattenSuzanne Brøgger ToneJan Lyderik
  Tangs saga. Bind 1-2

52
Similar entities

• Statistical based e.g. vector space model using
  tfidf weights

Entity 1
Entity 2
Shared elements
53
Statistisk baseret lighed mellem dokumenter

• Similaritetsmål mellem dokumenter bruges til
  identifikation af relationer. Links etableres på
  basis af tærskelværdier.
• Den mest udbredte teknik er anvendelsen af tf x
  idf vægtning i relation til vektorrumsmodellen.
• Dette vil typisk indebære flg.
• Betydende features/ord identificeres
  (stopordsliste)
• Ordene vægtes baseret på tfxidf og evt.
  positionelle parametre
• Similariteten mellem Dok.1 og alle andre
  dokumenter i basen beregnes udfra given algoritme
  (f.eks. som cosinus mellem to vektorer
  (dokumenter))
• Similaritetsværdier overstigende given
  grænseværdi gt etablering af link

54
Eksempel

• TermA, optræder i dok1 ift dok2
• tfA 5, men termen optræder en gang som Metadata
  og en gang i overskrift gt tfA7 (3X1 2X2)
  (Croft anvender tf ift. hyppigste term i dok)
• Idf 1/1000 (Croft anvender log. N/n)
• Termvægt for termA mellem dok1 og dok2 7/1000
• Samtlige termvægte indregnes i algoritme. F.eks.
  Udregning af cosinus.

55
(No Transcript)
56
Similarity example
Similar pages
57
(No Transcript)
58
Third part pointers

• Co-citations
• Similar to others who has lent this book, has
  lent these materials
• But from an author/domain perspective
• Example from Citeseer -gt

59
Citeseer example

• Abstract Latent Semantic Indexing (LSI) is a
  technique for representing documents, queries,
  and terms as vectors in a multidimensional
  real-valued space. The representations are
  approximations to the original term space
  encoding, and are found using the matrix
  technique of Singular Value Decomposition. In
  comparison, Multidimensional Scaling (MDS) is a
  class of data analysis techniques for
  representing data points as points in a
  multidimensional real-valued space. The objects
  are represented so that... (Update)
• Cited by More
• Automated Modeling and Nonlinear Axis Scaling -
  Leejay Wu (2005) (Correct)
• Similar documents (at the sentence level)
• 8.5 Optimizing Ranking Functions A
  Connectionist Approach to.. - Bartell (1994)
  (Correct)
• Active bibliography (related documents) More
  All
• 0.2 A Survey of Information Retrieval and
  Filtering Methods - Faloutsos, Oard (1996)
  (Correct)
• 0.2 Document Space Models Using Latent
  Semantic Analysis - Gotoh, Renals (1997)
  (Correct)
• 0.2 Approximating Matrix Multiplication for
  Pattern Recognition Tasks - Cohen, Lewis (1997)
  (Correct)
• Similar documents based on text More All
• 0.5 Chapter 15 Getting Better Results With
  Latent Semantic Indexing - Nakov (2000)
  (Correct)
• 0.4 Image Retrieval using Latent Semantic
  Indexing - Pecenovic (1997) (Correct)
• 0.4 On the Use of Singular Value Decomposition
  for Text Retrieval - Husbands, Simon, Ding (2000)
  (Correct)
• Related documents from co-citation More All

Co-citation threshold
60
Conclusion and perspectives Designing OPACs and
integrated search tools according to relations

• A lot of possibilities lots of types of
  relationships to display and utilize in different
  ways
• Bibliographic families Shared characteristics
  Whole-part and other bibliographic relations
• Similarity (statistical) Co citations User
  defined (co use)
• A.o.
• Need for carefull design of system features/link
  structures and a lot of testing (not only
  emploing user satisfaction but essentially
  improved search results)
• In other words Pick the functionalities that
  works for the user not the ones you like or are
  familiar with
• In general, we lack large scale user
  investigations