A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen...

12
A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien Lee Jane Liu, De-Nian Yang Research Center for Information Technology Innovation & Institute of Information Science Academia Sinica

Transcript of A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen...

Page 1: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

A Platform for Location Aware Service

-- with human computation

Ling-Jyh Chen, Meng Chang ChenMing-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien LeeJane Liu, De-Nian Yang

Research Center for Information Technology Innovation& Institute of Information ScienceAcademia Sinica

Page 2: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Itinerary Recommendation System with Human Computation

2

Itinerary Recommendation System with Human Computation

Page 3: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Other Travel Apps in Handheld Devices

• Nearby Spots (LBS)– Android Pocket Journey

– Android Wikitude

– Garmine

• Static Travel Routes (ebook)– Garmine

– MioMap

– TomTom

3

Page 4: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Our Recommendation System

• A data mining approach with GPS to provide “route” or “itinerary” based LBS

• Main characteristics– Personalization

– Human computation

– Quick and Dynamic Mining

Main Concept

System Architecture

GPRS/WiFi/…

Page 5: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Recommendation ServerRecommendation Server

• Kernel Modules– MSTravel

• An mining algorithm to discover user movement An mining algorithm to discover user movement regularity (itinerant patterns) from itinerary dataset regularity (itinerant patterns) from itinerary dataset

– Weight Grade• A grading function to select top-k suitable patterns

for rendering

User Travel LogUser Travel Log RecommendationRecommendation

Recommendation ServerRecommendation Server

Page 6: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Itinerant Patterns MiningItinerant Patterns Mining

• Spirits – Inherent from association rule mining and sequence

pattern mining

• Modeling itinerant pattern as a tuple (V, C, R)– V is an unordered set of visited scenic spots– C is the current location– R is an ordered sequence of recommended scenic spots– EX: (AB, C, DEF)

• Definition of Itinerant Patterns Mining ProblemGiven aGiven a itinerary dataset, discovering all itinerant patterns itinerary dataset, discovering all itinerant patterns with popularity minimal ratio ≧with popularity minimal ratio ≧ rrminmin and frequency minimal ≧ and frequency minimal ≧threshold threshold ttminmin

Page 7: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Itinerant patterns vs. Sequential PatternsItinerant patterns vs. Sequential Patterns

• Itinerant patternsItinerant patterns– Prune irrelevant sequences

• Render local characteristics• Provide more knowledge for

recommendation • Low computing complexity

• Sequential PatternsSequential Patterns– Consider all sequences

• Blur important local characteristics

• High computing complexity

popularity popularity of (A,B,C) = # of itineraries contain *A*B*C* / # of itineraries contain *A*B*=|{2}|/|{2,3}|= 0.5

frequencyfrequency of (AB,C, EG)= # of itineraries contain *A*C* or *B*C*=|{1, 2}| =2

Ex.

Page 8: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Itinerant Patterns Mining Algorithm -- MSTravelItinerant Patterns Mining Algorithm -- MSTravel

• A Recursive Approach– Explore k-1 Itinerant patterns– k-candidate Generation– Popularity-Testing

• against the minimal ratio and minimal threshold

– Redundancy-Elimination• prunes shorter itinerant patterns that are covered by the new

discovered ones

• Advantages of MStravel– Prune irrelevant itineraries → reduce DB scan

– Utilize apriori property in candidate generation → reduce the amount of comparisons in testing

Page 9: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Itinerant Patterns RecommendationItinerant Patterns Recommendation

• Multiple relevant patterns

Which one to recommend?

How to rank the patterns?

or

Page 10: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Weight Grading

• Possible recommendation strategies – Random k patterns, most popular k patterns,

longest k patterns, …

• Our solution– Weighting to obtain top-k patterns

• Consider popularity and frequency of a patterns • Consider similarity of a pattern and user’s visited

spots F = wF = w11 * popularity + w * popularity + w22 * frequency * frequency + +

ww33 * Jaccard (V, S) +w * Jaccard (V, S) +w44 * Jaccard (R, S) * Jaccard (R, S)

Advantage: K can be designated according to various applications

w1~w4: weight factors, S: user’s visited spots

Page 11: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Client-end Design

• Web-based UI – Incorporated with Google Map

– Simple operation

– User friendly

• Mapping of a geographic coordinates (x,y) and a scenic spot – Positioning accuracy and multiple hot spots in the same

location → not easy to identify user’s visited spot

– List top-k near by spots or list top-k

popular spots → users select manually

Page 12: A Platform for Location Aware Service -- with human computation Ling-Jyh Chen, Meng Chang Chen Ming-Syan Chen, Sheng-Wei Chen, Jan-Ming Ho, Wang-Chien.

Client: Web-based