Building and Evaluating a Location-Based Service Recommendation System with a Preference Adjustment mechanism Expert Systems with Applications, vol. 36,

Building and Evaluating a Location-Based Service

Recommendation System with a Prefer-ence

Adjustment mechanism

Expert Systems with Applications, vol. 36, no. 2, pp. 3543-3554, 2009

M.-H. Kuo et al.

Introduction

• Mobile commerce(M-Commerce) is developing trend due to successful experience of E-Commerce(EC)

• Most significant difference between M-Commerce and EC

mobility feature of mobile device

• Studies focusing on location data are created characterized as location-based service(LBS)

2

Previous Works

• Basic theme is based on that relevant information changes according to the location of mobile customers (Chen, 2002)

• The service have been adopted for various purpose– L-PRS: a location-based personalized recommender system (Kim, Song, & Yang, 2003)

• The key for LBS is the development of interface design and ability to provide correct and real-time content

– A user-oriented contents recommendation system in peer-to-peer architecture (Kim, Kim, & Cho, 2008)

Preference adjustment is necessary to recommendation system

3

Goal of this research

• Establishing a location-based information recommend system

– Integrating geographical location and personnel prefer-ence

• Developing and measuring a personalized prototype system based on location based service recommenda-tion model(LBSRM)

– Recommend hotel information

• Designing an experimental method for effectiveness of

preference adjustment– Long-term preference– Short-term preference

4

Recommendation model for LBS

5

Location-based database

• is an attribute set of LBS items

• Dataset : including dynamic and static attribute– Dynamic : - numeric type, catalogic type

6

User preference database

• Include user static data and dynamic data• : preference cluster of location based information of

different users

7

User history database

• Database record historical items user has selected

• Including contents– Mobile device identification code– System recommended items– User actually selected items– Content of each item

8

Recommendation module

• Area data filtering

– D : search area, (Xu, Yu) : user location

• Information grouping

9

Recommendation score calculating

• Score of numeric type

• Score of catalogic type

• Total recommendation score

10

r : rate of time-discountp : total number of recom- mendation itemsq : number of history record

Preference adjustment : Short-term ad-justment

• Use the difference of the recommendation score

• To prevent user preference goes to 0

• : recommendation score when the user selects item C• : recommendation score when the most recommended

information is item A

11

Preference adjustment : Long-term ad-justment

• Use Bayes’s decision procedure

• Add time discount rate ( r )

– : group number of items : the number of items oc-curred

: user selecting item of : system recommendation item : the sum of recommenda-

tion numbers

12

Development of prototype system

• User location based hotel recommendation system

13

Development of prototype system

• User location based hotel recommendation system

14

Development of prototype system

• User location based hotel recommendation system

15

Development of prototype system

• User registration

16

C.P number

3 preference of att. ofnumeric type

(distance, pricing, ser-vice)

Att. of catalogic type

multiple-choice

Development of prototype system

• Recommendation simulation system

17

Electronic map

C.P Inter-face

Recommendation simulation system• User location selection

– Randomly pick a location in order to simulate mobile environment

– Search all hotels with in search range– Using Euclidean Distance

• Recommendation step– Dived into three groups

• distance(D), Price(P) and service(S)– Take recommendation

• Calculation of the scores and display at the table• User can click “details’ and see further discription

• Preference adjustment step– Get the user feedback– Recommendation success or – Preference adjustment is then undertaken

18

Evaluation of prototype system

19

Evaluation of prototype system

• Satisfaction of system recommendation– Precision of recommendation

– On-line questionnaire is generated for the first 10 items• 35 registrants. Each respondent conducted for six times• Respondent : MBA student enrolled in National Defense Uni-

versity ,Taiwan

20

Evaluation of prototype system

21

Evaluation of prototype system

• The efficiency of preference adjustment– Short-term preference

22

Evaluation: Efficiency of preference ad-justment

• Short-term preference

23

Lack of sta-bility

Evaluation: Efficiency of preference ad-justment

• Long-term preference

24

Evaluation: Efficiency of preference ad-justment

• Long-term preference

25

1. The system reaction to preference ad-justment is immediate if without his-tory data

2. Latter adjustment(i.e., the more history data), the more needed number of adjustment

3. The lower the discount rate, the faster the learning adjustment speed

Evaluation: Efficiency of preference ad-justment

26

Evaluation: Efficiency of preference ad-justment

27

Evaluation: Efficiency of preference ad-justment

28

Conclusion

• Proposed a model combines LBS and information recom-mendation

• Designed prototype system of preference adjustment

• Evaluated the effectiveness of the LBSRM as adopt the on-line questionnaires

– Proved that the expected recommendation effect is to be achieved

• Evaluated the efficiency of the preference adjustment

• Regard to the adjusting ability, long-term preference can reach better result

• Number of preference adjustment could be determined by changing the weight of recent preference(i.e., dynamic method of time-dicount rate)

29