A Behavioral Pattern Adapted to Individual for Providing Ubiquitous Service in Intelligent Space

1
A Behavioral Pattern Adapted to Individual for
Providing Ubiquitous Servicein Intelligent Space
Hiroyuki Yamahara Hideyuki Takada Hiromitsu
Shimakawa
yama_at_de.is.ritsumei.ac.jp
simakawa_at_cs.ritsumei.ac.jp
htakada_at_cs.ritsumei.ac.jp
Ritsumeikan University
1-1-1 Nojihigashi Kusatsu Shiga 525-8577 Japan
WSEAS CIMMACS2006 in Venice
Nov. 20, 2006
2
Assist user activity in ubiquitous environment

• There are lots of researches to assist user
  activity in a ubiquitous environment.
• It is useful for solitary old people and two
  income family, who want help from outsiders.
• It is desirable to be proactively provided
  services to assist user activity according to
  user intention.

3
Proactive services according to user behavior

• Our research aim to provide proactive services
  according to user behavior which indicates user
  intention.
• It improves user amenity.
• It brings relief and safety to a user by
  preventing danger in advance.
• Examples of service a scene of going out
• To warn that a gas valve is not close
• To notify that a user left something behind
• To call an elevator to living floor beforehand

4
When does a user need assistance ?

• A user does not need assistance in every scene of
  daily life.
• It is desirable for a user to be assisted in a
  scene in which users mode significantly changes.
• Scenes of going out, coming home, getting up,
  going to bed
• Suppose you have gone out without closing a gas
  valve

5
Provide service focusing on change of mode

• We aim to
• detect user behavior in a scene in which users
  mode significantly changes, and
• provide services proactively before users mode
  has changed.
• Improvement of user amenity in daily life
• Proactive prevention of danger

6
Cycle of detection of user behavior
Time Passage
Some behavior logs are collected as samples.
behavior log
A behavioral pattern is created with sample
behavior logs.
behavior log

behavioral pattern
behavior log
User behavior is detected by matching current
behavior log with a behavioral pattern.
behavior log
7
Existing behavior recognition method

• Existing methods represent a behavioral pattern
  with a probabilistic model such as Hidden Markov
  Model.
• It represents user behavior as a series of state
  transition between two state.

State 1
State 2
State 3
State 4
8
Problems in existing method

• They need a lot of sample behavior logs to create
  a behavioral pattern.
• It cannot realize early start of service to a
  user.
• They cannot recognize user behavior which
  contains exceptionally rare actions.
• Exceptional rare actions are often weaved into
  actual user behavior in daily life.

9
Tagged World

• An intelligent space to identify targets of user
  operation in daily life.
• Users behavior logs are collected with
  close-range RFID system.
• history of tag-ID of objects a user touched and
  its time stamp

Tagged World
10
Individual habit represented by objects

• A user has habitual actions individually in a
  scene in which his mode changes.
• Suppose a scene of going out
• pick up a wallet, go to the toilet, and so on
• Everyone often touches same objects in a scene.
• Individual habit is represented by kind of
  objects a user touched and order of them.
• The Tagged World creates an individual behavioral
  pattern from individual habit.

11
How to create a behavioral pattern
case_ID 1 100000055 pants
hanger 100000017 lavatory faucet 100000018
lavatory cup 100000020 toothbrush 100000019
toothpaste 100000020 toothbrush 100000018
lavatory cup 100000020 toothbrush 100000018
lavatory cup 100000017 lavatory faucet 100000068
cell phone [phone] pass case 100000065 wrist
watch 100000050 bag
ordered pair count pants
hanger ? lavatory faucet 9 pants
hanger ? lavatory cup 18 pants
hanger ? toothbrush 10 pants
hanger ? toothpaste 10


(1)Collect behavior logs
(3) Count occurrence
(2) Enumerate ordered pairs
(4) Ordered pairs more than the threshold are
extracted
p1 pants hanger ? lavatory cup p2 pants hanger
? cell phone p3 cell phone ? milk carton p4
cell phone ? bag p5 wrist watch ? bag
Ordered pair An order between two objects
when a user touches an object B after
he touches object A. ?Two objects must not be
successive.
Behavioral pattern represented by a set of
ordered pairs
Some cases of behavior logs
12
How to match with a behavioral pattern
Behavior log to be matched
Behavioral pattern represented by a set of
ordered pairs
p1 pants hanger ? lavatory cup p2 pants
hanger ? cell phone p3 cell phone ? milk
carton p4 cell phone ? bag p5 wrist watch ?
bag
200000613 wardrobe 200000597 hanger 200000613
wardrobe 100000068 cell phone [phone] pants
hanger 100000017 lavatory faucet 100000018
lavatory cup 100000020 toothbrush 100000019
toothpaste 100000020 toothbrush 100000018
lavatory cup 100000020 toothbrush 100000018
lavatory cup 100000020 toothbrush 100000018
lavatory cup 100000017 lavatory faucet 100000071
refrigerator 100000072 milk carton 100000071
refrigerator
calculate the characteristic point according to
the number of occurrence of each ordered pair
ordered pair number of occurrence
p1 pants hanger ? lavatory cup 4
p2 pants hanger ? cell phone 0
p3 cell phone ? milk carton 1

13
Features in the proposed method

• Representation of a behavioral pattern with not
  only successive two objects but also
  non-successive two objects in a behavior log
• Separation of probabilistic model from order
  check in matching with a behavioral pattern

• These features can flexibly characterize user
  behavior of each scene.
• Our method isn't affected by exceptional rare
  actions weaved into actual user behavior.

14
Experiment

• Target scene to detect
• Going out
• (before he has gone through the entrance door)
• Scenes of collected behavior logs
• as samples to create a behavioral pattern
• Going out
• as logs to be matched with the pattern
• Going out, Coming home, and so on.
• A behavioral pattern is created with 3 to 10
  sample behavior logs.
• with 15 experimental subjects

15
Recognition rate
Detection rate of going out
Exclusion rate except going out
The number of sample behavior logs
16
Discussion

• A behavioral pattern of going out
• detected 98 of user behavior of going out
• excluded 95 of user behavior in other scenes
• A behavioral pattern of coming home
• detected 93 of user behavior of coming home
• excluded 92 of user behavior in other scenes
• Less objects are touched in a scene of coming
  home compared to a scene of going out.
• There are less characteristics of user behavior.
• The experiment has proved our method can create a
  behavioral pattern with less than 10 behavior
  logs in a practical short time.

17
Conclusion

• The Tagged World aims to provide services
  proactively before users mode has changed, by
  detecting user behavior in a scene in which
  users mode significantly changes.
• Our method
• individualizes user behavior with order of
  objects a user touches in every scene, and
• starts service to a user by creating an
  individual behavioral pattern in a short time,
  and
• detects user behavior without being affected by
  rare order of users action.
• In the future, we will let a portable computer
  collaborate closely with living space information
  to provide more effective and appropriate
  services.