Machine Learning and Intention Recognition Systems with Self-Knowledge Acquisition

1
Machine Learning and Intention Recognition
Systems with Self-Knowledge Acquisition
H. Kanso, A. Elhore, C. Soulé-Dupuy, S. Tazi
IRIT, LAAS, Toulouse 1 University, France
ICHSL08, Toulouse, France
2
Outline

• Research topic
• Notion of intention
• Contribution
• Model of Intentional Structure
• Intention Recognition Systems
• knowledge acquisition in Pero System
• knowledge acquisition in Analyzer System
• Conclusion and Perspectives

3
Research topic

• What are we doing ?
• Model, represent and process intentions of the
  writing actions
• Define new structures of document Intentional
  Structure
• Why?
• Enable the access to documents in terms of
  authors intentions
• How?
• Specifying a conceptual model of intentions
• Coding intentions as Xml metadata inside
  documents
• Try to recognize these intentions (semi
  automatically)
• For what reason ?
• Understand human cognitive processes
• Contribute to the progress of this new area
• Have more funded projects,
• Have more interested students,
• Etc.

4
Notion of Intention

• The concept of intention is omnipresent in any
  human action and is particularly important in
  communication.
• It is a mental state of any actor who executes an
  action.
• We are interested exclusively by intentional
  actions, i.e. premeditated actions.

5
Intentions /documents

• Relations between segments and authors
  intentions define Structural Intentions of
  documents.
• Grosz and Sidner Theory Grosz 86
• Coherence of speech based on a set of intentions
  and a small set of relations between intentions.
• RST is a descriptive and functional theory of the
  textual organization based on the recognition of
  semantic relations between units of text.

6
Intentional Structure Model

• I (A, G, M, R)
• Where
• I represents the intention carried out by an
  action A
• A is the action that expresses what the author of
  the intention wants to do
• G represents the goal to achieve by performing
  the action
• M represents the means to express how the action
  is accomplished
• R represents the reason to express why the author
  chooses this action and for which reasons.
• indicate that the number of the acts composing
  the intention can be 0 or N.

7
What are we doing ? Model, represent and process
intentions of the writing actions Define new
structures of document Intentional Structure
Action
Why? Enable the access to documents in terms of
authors intentions
Goal
How? Specifying a conceptual model of
intentions Coding intentions as Xml metadata
inside documents Try to recognize these
intentions (semi automatically)
Means
For what reason ? Understand human cognitive
processes Contribute to the progress of this new
area Have more funded projects, Have more
interested students, Etc.
Reason
8
Intention Representation
9
Example
Segment A gardener digs a hole with the aim of
planting a tree by using a shovel, in order to
decorate his garden

• A digs (a hole)
• G Planting (a tree)
• M Using (a shovel)
• R Decorate (the garden)

10
Example (1/2)
G
R
M
11
Example(2/2)
12
Intention Recognition Systems

• Pero and Analyzer System are based on set of
  algorithms which facilitate the intentions
  research.
• Their principles closer to an expert of a domain.

13
Intention Recognition Systems

• Pero a planner implemented in Java to solve
  exercises in physics automatically in such a way
  that the students can make observations on the
  proof.
• Analyzer is dedicated in the information research
  of textual corpus.

14
knowledge acquisition in Pero System

• (a1, a2, A, G, M, R)
• a1 is the author of the action it is generally
  the system,
• a2 is the agent for which is intended the
  explanation of the action generally is the
  learning,
• G (Goal) is an act which expresses what the
  author wants to do by making the action
• M (Means) is an act which expresses the type of
  action achieved on the reasoning
• R (Reason) is an act which expresses with which
  concepts the author makes the action.

15
knowledge acquisition in Pero System
16
knowledge acquisition in Pero System
17
knowledge acquisition in Analyzer System
18
knowledge acquisition in Analyzer System
19
Analyzer Output
20
Conclusion and Perspectives

• We worked on the model of intentional structure
  to detect the intention of a written document.
• We presented two intention recognition systems
  (Pero and Analyzer), at the aim to detect the
  intention in scientific domain.
• We focused on the knowledge acquisition in these
  two systems, and how its update the knowledge
  base automatically by new acquired data.
• The use of ontologies in our case makes it
  possible to facilitate the intentions
  recognition.
• We see this work as a step towards creating a
  solution for automatic intention recognition in
  the textual corpus and to enhance the acquisition
  knowledge for enrichment the knowledge base.

21
Thank you!