personalizing threshold values on behavior detection with collaborative filtering

26
1 Personalizing Threshold Values on Behavior Detection with Collaborative Filtering

Upload: wacerone

Post on 13-Jun-2015

530 views

Category:

Technology


2 download

TRANSCRIPT

Page 1: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

1

Personalizing Threshold Values on Behavior Detection with Collaborative Filtering

Page 2: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

2

Reference

Hiroyuki Yamahara, Fumiko Harada, Hideyuki Takada,and Hiromitsu Shimakawa “Personalizing Threshold Values on Behavior Detection with Collaborative Filtering” Ubiquitous Intelligence and Computing 2008, LNCS 5061

Page 3: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Outline Introduction Behavior Detection in the Home Discussion on Setting of Threshold Values Dynamic Threshold Determination with

Collaborative Filtering Related Work Conclusion

3

Page 4: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Introduction A collaboration system for assisting users in

their homes, as an attempt for making intelligent environments

Collaboration with the user by environment may bring him comfort, relief and safety

proactive services high-level behaviors set threshold values, which are used for creating

a behavioral pattern and for matching online sensor data with the pattern

4

Page 5: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Behavior Detection in the Home

Detection of High-Level Bbehavior

Individual Habit in Touched Objects

Behavior Detection with Ordered Pairs

5

Page 6: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Detection of High-Level Behavior leaving the home, coming home, getting

up and going to bed consider that a high-level behavior is a

long behavior of around ten minutes characteristics of the highlevel behavior

vary with individual user

6

Page 7: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Individual Habit in Touched Objects most people often have habitual actions in

a habitual order, for not making omission of things to do, in situations such as leaving the home and going to bed

RFID tags

7

Page 8: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

8

Page 9: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

9

Page 10: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Behavior Detection with Ordered Pairs create a behavioral pattern represented by

a set of ordered pairs, which show the order relation among touched objects, with histories of touched objects as sample behavior logs

10

Page 11: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

How to create a behavioral pattern

11

Page 12: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Discussion on Setting of Threshold Values Difficulty of Setting Threshold Values Effect of Detection Threshold on

Recognition Accuracy of User Behavior Effect of Extraction Threshold on

Recognition Accuracy of User Behavior

12

Page 13: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Difficulty of Setting Threshold Values TPR shows the rate at which behavior logs in a

specific situation, which logs are referred to as true cases, are correctly detected with a behavioral pattern of the situation

TNR shows the rate at which behavior logs in situations other than the specific situation, which logs are referred to as false cases, are correctly neglected with the behavioral pattern of the situation

13

Page 14: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

14

Page 15: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

15

Page 16: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Effect of Detection Threshold on Recognition Accuracy of User Behavior

Comparing the recognition rates in tables from Table 1 to Table 4 with differences between common values and the best values of each subject in Table 5, it is apparent that the more differences bring lower recognition rate

16

Page 17: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

17

Page 18: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Effect of Extraction Threshold on Recognition Accuracy of User Behavior

18

Page 19: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Dynamic Threshold Determination with Collaborative Filtering Dynamic Determination of Thresholds

Suitable for Individuals Determination of Thresholds with Estimate

Values by Collaborative Filtering

19

Page 20: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Dynamic threshold determination with collaborative filtering

20

Page 21: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Dynamic Determination of Thresholds Suitable for Individuals

21

Page 22: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Determination of Thresholds with Estimate Values by Collaborative Filtering

22

Page 23: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Related Work

23

Page 24: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

Conclusion

Realize the collaboration by providing proactive services according to user behavior

To detect user behavior precisely, our detection method dynamically determines threshold values suitable for behavioral patterns of individuals with collaborative filtering.

Our future work is evaluation of the determination method in experiments.

24

Page 25: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

25

Page 26: Personalizing Threshold Values On Behavior Detection With Collaborative Filtering

三種過濾方式之比較

26