From Documents to Knowledge Models

1
From Documents to Knowledge Models

• Max Völkelvoelkel_at_fzi.de
• Forschungszentrum Informatik an der Universität
  Karlsruhe (TH)

2
Personal Knowledge Management

• Definition knowledge cues Haller
• any kind of symbol, pattern or artefact which
  evokes some knowledge in a persons mind, when
  viewed or used.
• Knowledge cues can be stored and retrieved on a
  computer while knowledge may or may not.
• Ok, in fact you store bits (signals)

3
What is a Document?
A team of 50 French researchers discussed
4
Definition Document

• A team of 50 French researchers could agree on
• Document as form
• Document as a container, which assembles and
  structures the content to make it easier for the
  reader to understand it.
• Document as sign
• Emphasize argumentative structure of the content.
  
• Document can be referenced ? acts as a sign for
  its meaning.
• Document as medium
• Reading contract intention or assumption of
  the author what will happen with the document.

5
Document (my definition) I/II

• A document consists of information atoms.
• An information atom is the smallest unit of
  content which can be interpreted without a
  documents context (but of course requiring
  background knowledge). For text, these atoms are
  single words.

• Packaging establishes a context
• Reference-ability reference to a published
  document can act as a placeholder for the content
  expressed within.
• Process metadata should be sent along
• such as authors, audience, goal

Document
Author, audience, goal
6
Document (my definition) II/II

• A document is a knowledge artefact consisting of
  several layers

Content Semantics

• content means something.
• Building upon logical and argumentative
  structure, the author encodes statements about a
  domain within the content.

• to convey its content to the reader.
• Argumentative structures appear on all scales. A
  typical structure is the Introduction - Related
  work Contribution - Conclusion-pattern of
  scientific articles. On smaller scales, patterns
  like claim-proof and question-answer are
  used.

Argumentative Structure
Logical Structure

• can reference smaller parts within a document
• i.e. paragraphs, headlines, footnotes, citations,
  and title

Visual Structure

• guides the reader informally
• type-setting (i.e. bold, italics, different font
  styles and size), placement of figures, pages
  carries additional information

Linearity

• defined order
• for navigating through all information items

7
Ted Nelson

• I propose a different document agenda
• I believe we need new electronic documents which
  are transparent, public, principled, and freed
  from the traditions of hierarchy and paper.

8
What do people want?
?Why?
9
What is a Wiki? Whats new compared to CMS?

• Easy Contribution ? shorter time-to-publication
• Wiki pages can be created and edited by any user
  quickly and easily
• Easy Writing
• Simple text formatting without the need to learn
  HTML ? Wiki Syntax
• Easy Linking
• Automatic linking converts written names of
  pages, images and websites to links
• Recent Changes
• See what has happened Awareness
• Diff function shows the latest changes
• Easily check whether changes are ok
• Fulltext search for page titles and text
• Backlink function shows which pages link to the
  current page
• Find the context of this page
• Directly link deep into a wiki using readable
  names

Wikis were the first deployed, collaborative hyper
text authoring environments ? People want more
links
10
My definition based on OMG metamodel MOF
What is a Model? Typed entities and typed
relations
TypeA2
TypeB2
Type C2
(Meta-)Modelling
TypeA1
TypeB1
TypeC1
Modelling
EntityX
EntityY
Real world from theviewpoint of the individual
ArtifactX
ArtifactY
11
What is a Knowledge Model?
Document Ontology Knowledge Model
Information atoms Text (paragraphs, images, multimedia resources) Concepts Items (text, images, other binary resources)
- Text Short (headlines) and longer (paragraphs) Short labels Anything from short labels to structured documents
Order Strict linear order Yes, may be partial and have cycles
Hierarchy Yes (chapters, sections, paragraphs, sentences) Yes Yes, may be partial and have cycles
Annotations Yes (footnotes) Yes Yes
- Tagging (annotation with keywords) Yes
- Typing (inc. Inferencing) Yes Yes
Hyperlinks Yes (internal references and external citations) Yes, dont have to occur inside text
Visual layout Yes
12
From Documents to Knowledge Models

• From analogue to digital documents
• smaller content granularity
• more interconnected content
• more explicit structures.

• ? Knowledge models
• very small information atoms, such as single
  words
• Richly connected items
• explicit semantics for the links.

• Definition
• A knowledge model is a superset of
  documents and formal ontologies.
• Annotated documents, stored together with their
  annotations, can be seen as a knowledge
  model.

13
What is a CDS? Conceptual Data Structures
M. Völkel and H. Haller Conceptual Data
Structures (CDS) - Towards an Ontology for
Semi-Formal Articulation of Personal Knowledge
In Proc. of the 14th International Conference on
Conceptual Structures 2006. Aalborg University -
Denmark, July 2006.
14
What is a CDS-based Knowledge Model?

• A set of addressable items (text, images, maybe
  even multimedia elements)
• Relations between items, classified in four types
• Source/target the generic, directed hyperlink
  link
• Before/after ordering relations, linear
  navigation
• Context/detail hierarchical relations, document
  and concept hierarchies
• Annotation/annotationMember annotations, to
  give the ability to type items and relations,
  items are used as types ? meta-modeling
• Knowledge models must be able to capture
  work-in-progress
• CDS is not strict, you can have cycles, untyped
  items, paradox ordering,

15
CDS A Hierarchy of Relations
Legend
Undirected Relation related/related
informal
Relation Typerelation/inverse
Equivalency equivalent
Directed Linking source/target
Labelled Links /-inverse
Order before/after
Hierarchy detail/context
Annotation annotation/annotationMember
Subclassing is-a/superclass-of
Taskpriority
Tagging tag/tagMember
Documentorder
Instantiation type/instance
formal
16
Motivation
17
Examples for Knowledge Models

• Fiction Writing

18
How does Writing/Reading works?

• Writing / Sending
• Write down ideas
• Group them
• Structure them
• Add argumentation structures
• Add references to literature
• Link pieces in a first draft
• Add introduction and conclusion
• Repeat until coherent flow
• Publish document

• Reading / Recieving
• Visualise the structure graphically
• Connect new structures with existing own
  structures

Von der Idee zum Text Esselborn 2004
19
The tool chains break

• Create a new slide show out of three old
  presentation plus one from your colleague
• Why not have the content in smaller, more logical
  chunks?
• Re-use the motivation part of an old paper for a
  new one
• If you find a mis-spelling, why have to fix it
  twice?
• Search a stack of paper notes with good ideas
• Why are those not in your computer?
• Search email archives to find out what the
  high-level architecture for the new
  authentication system is
• Why not browse your PKM and see the relations?

20
Technological Developments

• ? accelerated distribution by many orders of
  magnitude
• ? lower costs

Analog ?
Digital
Communicationspeed
internet
printing press
cost
written language
time
21
Cost of Communication Data transmission is
cheap now

• Total cost of communication to send content to n
  people
• choosing relevant parts of the personal
  model encoding of model parts in document
  parts order document parts strictly
  linear/hierarchical n ( data transmission
  linear reading of the document
  decoding of model parts from document parts
  creating a networked model out of
  model parts integrate new model to
  existing model )

22
Cost of Communication Where can we save, if n
is small?

• Total cost of communication to send content to n
  people
• choosing relevant parts of the personal
  model encoding of model parts in document
  parts order document parts strictly
  linear/hierarchical n ( data transmission
  linear reading of the document
  decoding of model parts from document parts
  creating a networked model out of
  model parts integrate new model to
  existing model )

23
Cost of Communication

• Total cost of communication to send content to n
  people
• choosing relevant parts of the personal
  model encoding of model parts in document
  parts order document parts strictly
  linear/hierarchical n ( data transmission
  linear reading of the document
  decoding of model parts from document parts
  creating a networked model out of
  model parts integrate new model to
  existing model )

24
Current process culture is document-centric

• Sender

Recipient(s)
Cost
25
Ideal process - What if not documents, but
knowledge models would be exchanged between
people?

• Sender

Recipient(s)
Cost
26
Realistic (improved) process use both

• Sender

Recipient(s)
Cost
27
Information Management Problems ? Solution
Knowledge Models

• Under-utilisation of the interlinked nature of
  information Oren? fine-granular nature of
  knowledge models allows for precise and effective
  linking and browsing
• People have problems in using strict hierarchies
  Oren? classification methods like tagging and
  non-strict taxonomies
• Keep the context Oren ? networked nature of a
  knowledge model is more suited to represent
  contextual links than a set of documents
• Granularity ? Represent more than the content of
  just one document

28
When to use Knowledge Models?
Fixed domain

• Use domain specific tools languages
• Standardised representation formalisms
• Established data exchange processes

Open domain- or Multiple domains

• Use personal knowledge models
• Unstructured, semi-structured, semi-formal and
  formal parts
• Ad-hoc formalisation
• Cheaper to create, easier to integrate
• Use Documents
• Costly to create
• Cheap to read ? sometimes the best solution
• Hard to integrate

Myself! My TeamMy Community
Broad audience
29
Related Work in Semantic Authoring

• Initial ideas - although that term was not used
  - can be found already in V. Bush and D.
  Engelbart
• ABCDE Format from Anita de Waard
• Semantically annotated Latex (SALT) by Tudor
  Groza
• Systems allowing end-users to construct
  ontologies out of their linked information
  objects.
• L. Ludwig sees redundancy within and among
  documents as a hurdle to efficient information
  usage. Traditional notion of a document is
  replaced by virtual documents, which render parts
  of the knowledge base as an interactive tree.
• Bernstein describes TinderBox, a "personal
  content management assistant", which offers
  sophisticated HTML generation via templates.
• Gnowsis system by Sauermann allows to link
  desktop objects, integrates with wiki
• iMapping semantic concept maps by Haller
• Same direction in the fields of semantic desktop
  and semantic wiki
• Semantic Web Content Repository (swecr)

30
Conclusion
ContactMax Völkel, voelkel_at_fzi.de
Thank You very muchfor Your attention

• Documents
• Document-centered culture is a costly legacy
  artefact and bottleneck for our society
• Personal knowledge models
• Superset of documents and ontologies
• Integrate with the semantic desktop
• Make knowledge worker happier and more productive

• Authoring is the bottleneck
• We should bring the power of modeling to the
  end-user
• Dont break the tool chain
• Focus on work-in-progress