Transcript Design of Intelligent Interface

Design of Intelligent Human Interface
- Brief Summary of Research Activities -
Yasufumi Takama
Tokyo Institute of Technology
（Currently: Tokyo Metropolitan Institute of Technology)
PREST, Japan Science and Technology
Corporation
1/16
VSA 1
Characteristics of VSA

Multi-modal Interface



Suitable for users with little knowledge of
computers


Face-to-face interaction & speech dialog
Connected with WWW browser
Applicable to facility guidance systems
Dialog management system (DMS) with
learning module

Acquisition of cooperative answering strategy
2/16
VSA 2
DMS with Learning Module

Cooperative answer depends on both user’s
knowledge & system’s location




E.g. “Where is the library?”
Tell the way from system’s location to the library
Use the landmark that user seems to know
2 types of knowledge acquisition
 Overlay-type user-model to estimate the
knowledge state of the present user
 Reinforcement learning to acquire locationspecific appropriate rules
3/16
FISH VIEW 1
Document Ordering while Reading is
an Effective Way of Dealing with Vast
Collection of Documents
Making
diagrams
Reading
documents
Viewpoint
extraction
Viewpoint-based support
by Fisheye Matching
Retrieving
new documents
4/16
FISH VIEW 2
Extracted Concepts as Viewpoints

From medical news
ID

Heading Information
Words (stemmed)
3f98b3
value of health
diseas,sickne,health,etc.
444506
component of living body
protei,immuno,choles,dna
30f6da
internal organs
eye,heart,lung,knee,etc.
3f969e
disease
syndro,aids,cancer,cold,etc.
44479c
medical supplies
drug,medici,laxati,acid,etc.
30f6f7
medical instruments
bandag,cathet,glasse,etc.
From cinema reviews

Categories of action, SF, monster, etc.
5/16
Query Network 1
Query Network:
Plastic Clustering for Visualization


Clustering-based visualization applied to
WWW-IR
Categorization (static, i.e. user independent)
Plastic clustering
(gradual formation
through a series of IR)

Naive clustering (one-shot, i.e. no inheritance)

K-means, STC, etc.
6/16
Query Network 2
Analogy to Immune Network



Antibodies generated for
antigens
Activated antibody can
be sensitive to next
invasion
Various antibodies in
plastic structure



Clusters generated for
retrieved documents
Effective cluster can be
reused for subsequent
retrieval
Various clusters in
plastic structure
Activation value calculation
based on differential equation of
mathematical biology
7/16