Evaluation of Online Itinerary Planner & Investigation of Possible Enhancement Features

Transcription

1 Evaluation of Online Itinerary Planner & Investigation of Possible Enhancement Features Ho ming Tam & L.S.C. Pun Cheng Department of Land Surveying and Geo Informatics, HK PolyU JIC TDHM GIS, Hong Kong, May, th May, 2010

2 Initiative Tourists need to plan their itinerary well ahead their arrival Much effort spent on looking for interesting places to visit, but Even more effort spent on transportation plans Abundant yet fragmented transport information Utilizing existing information & resources 2

3 Transport Information Public Transport Enquiry Service (PTES) Transport Department, HKSAR Government, & Dept of Land Surveying & Geo Informatics, HKPolyU Point To Point transport route suggestions Similar services: Transport Info by NSW Gov t, Australia Transport Direct by Atos Origin, UK 3

4 PTES Search Map Search Text Search HKSAR Government HKSAR Government 4

5 PTES Result Presentation HKSAR Government 5

6 Attraction Information Tourist spots: Lonely Planet Dining: OpenRice.com OpenRice.com 6

7 What if more than 2 places? THE PEAK TIAN TAN BUDDHA WONG TAI SIN TEMPLE Wikipedia Robert Lai LAM TSUEN WISHING TREE An optimal sequence? Efficient transport route? Pick the least cost (Money / Time)? 7

8 Itinerary planner as a solution Examples discussed call for the presence of an itinerary planner Well developed and complex transport network makes every pair of Point To Point travel become possible Unless there are two isolate/separate sets of transport network, there must be a solution from one place to another 8

9 Objectives Provide scheduling solution & optimize it Save time for more tourist spots Provide transport information Bridge the gap among different sources of information (both transport and scenic spot) Efficient algorithm / approximation of Travelling Salesman Problem (No efficient algorithm to solve exactly) 9

10 Scope of Study Disneyland Fung Ying Seen Koon Lam Tsuen Wishing Tree Tian Tai Buddha Statue Avenue of Stars Clock Tower Golden Computer Arcade Lingnan Garden Temple Street Wong Tai Sin Temple Golden Bauhinia Square (Wan Chai) Lan Kwai Fong Legislative Council Building Repulse Bay Stanley The Peak ChinaTouristMaps.com 10

11 Greedy Algorithm & Heuristics Search Approximation algorithms Solutions needed in a short time Greedy algorithm (Constructive heuristics) Nearest Neighbour algorithm (NN algorithm) Solution for random case = 1.25 x shortest path K Opt Heuristic (Iterative improvement) Develop a 3 spot window which keeps iterating the sequence of the 3 spot Looking for the existence of a shorter alternative 11

12 Data Preparation Point Of Interest Data (POI) Lonely Planet Every single Hotel / Tourist spots as a Point Of Interest (POI) C i,j Cost (Price or Time) incurred when travel from i to j, where i <> j Destination 1 2 n TRANSPORT Data Public Transport Enquiry Service ( 1 2 Origin C 1,2 C 1,n C 2,1 C 2,n n C n,1 C n,2 12

13 Entity Relationship Diagram Spot (POIs) Hotels & Tourist Spots Journey Cost data Total travelling time SubJourney Individual leg Detail leg info. 13

14 Assumptions 1. ONE and ONLY ONE route between every two spots (Asymmetric graph) 2. All the transport routes are available all day long as day time transport A schematic representation of a network with 4 POIs, e.g. The two arrows connecting 1 and 2 corresponds the In and Out bound journeys of the pair. 14

15 A two way way journey Asymmetric In bound and Out bound journey are different: Pick Up & Drop Off Stops Sequence of interchange (Bus Train, Train Bus) Fare Time taken (Out bound may take short time than In bound) 15

16 A two way way journey Asymmetric Example case: PolyU The Peak 1. PolyU The Peak (In bound) 2. The Peak PolyU (Out bound) HKSAR Government HKSAR Government 16

17 Scheduling Algorithm Prepare STARTING SOLUTION System picks up the list of spot selected by the user Default starting point: Hotel selected by user Keep visiting the nearest spots until the end of the trip (Nearest Neighbour algorithm) Improve STARTING SOLUTION BY ITERATIVE K opt heuristic (k value = 3) 17

18 K opt Heuristics Improvement Window Fixing the nodes before and after the improvement window, lowest cost will be adopted. Fixed nodes Once computed, 6 permutations the (3! = improvement 6) window will be shifted forward. Fixed nodes 6 permutations (3! = 6) 18

19 K opt Heuristics Improvement Window Improvement window (3 Opt Heuristics) 19

20 K opt Heuristics Improvement Window Window starts at node 2 and move until its end reaches node n 1 START n n END 1 20

21 K opt Heuristics Significance Hotel: 11. The Peninsula Tourist spots: 1. Legislative Council Building 3. Golden Bauhinia Square 6. Clock Tower 7. Temple Street BEFORE 17. Lan Kwai Fong AFTER 21

22 Intelligence One EYE takes care of 7 things Plan the trip for you Plan i itinerary Planner

23 Demonstration HOTEL: The Peninsula START TIME: 10:00 END TIME: 20:00 23

24

25 Google Map 25

26 Prompt of Overtime May require more than 1 day to complete System will prompt user Transport route will be altered accordingly 26

27 Time Restriction on Tourist Spot Breakfast Time? OpenRice.com Google Map 27

28 Itinerary with Time Restricted Spots Golden Bauhinia Square (Wan Chai) 07:00 09:00 Golden Computer Arcade 12:00 22:00 Temple Street 19:00 23:00 Lan Kwai Fong 20:00 22:00 28

29 Itinerary with Time Restricted Spots Possible Solutions: Iterative (Brute Force Search) Time Slot Fitting Approach 29

30 Iterative Solution Exhaustive Search (N!) Do NOT consider the time issue Compare every solution to see if the spots concerned fall within the time slot (Comparison of some odd solutions) Expensive transactions (Computation power limited with mass request) 30

31 Time Slot Fitting Is this what you did without a computer? Tourist spots (General); hereinafter as non TR spot Tourist spots (Time Restricted); hereinafter as TR spot 31

32 Time Slot Fitting: Algorithm (Part 1 of 2) For(every TR spot in itinerary){ if(tr spot is time restricted && not complete lie within user time slot){ remove TR spot from itinerary } }//Spots in itinerary should all be achievable. CASE 1 Available time User activity time Available time A B AM Morning Tea Other time of the day A B Time for travel TO morning tea Time spent during morning tea CASE 2 Available time Available time User activity time B A Lan Kwai Fong A Time for transport leaving for next spot Other time of the day PM B Time spent during LanKwaiFong

33 Time Slot Fitting: Algorithm (Part 2 of 2) While(unarranged spot > 0){ while(tr Spot > 0){ if(tr > currenttime){ Insert 1st TR spot (or TR spot of same district) Put it as far end as possible while(some time in between available time & TR){ Insert non-tr spot in between TR and start time } shift EARLIER whole bunch to stick to start time }else{ while(some time in between TR end & end of day ){ Insert non-tr spot in between TR and start time Put current time = start time of next day} } } Add other spots using NN } Perform Heuristic starting from spot(last TR + 1) While(putting spots after last TR){ if(time visit next spot > user specified end time){ current time = start time of next day } Return result

34 Time Slot Fitting: Algorithm A time line of the user defined time slot 34

35 Insertion of a TR with nearest start time To the far right end If there is some time in between, look for possible non TR spot (not possible Here) else shift the whole bunch earlier 35

36 Insertion of a TR spot in the same district with previous TR with earliest start time To the far right end If there is some time in between, look for possible non TR spot (not possible Here) else shift the whole bunch earlier 36

37 If(there is time between the two bunch){ } insert non TR spot of different district using NN if (time not sufficient){ extend the time stay in non TR spot OR suggest some other spots of same district } shift left 37

38 while(there is time between end time of last TR & user end time){ Repeat steps in 2 slides before Insert non TR spot If(time not sufficient){ Put the transport back to the hotel } } Start another day, repeat the whole cycle 38

39 Difficulties in implementation Both solutions (Iterative and Time Slot Fitting) requires A huge number of comparison Repetitive insertion and deletion Sorting Extreme number of transaction 39

40 Difficulties in implementation Handling large number of simultaneous requests Stress Test Data Preparation Manual data retrieval Data retrieved well ahead of request Frequent update of PTES Pre determined transport means 40

41 Further Enhancement Look for the ability to reduce the problem size for the time constrained case Interactive drag and drop scheduling User may choose to stick with a particular kind of transport (whenever possible), avoid 1 and only 1 solution Generate transport data automatically Back end computation of lists of popular tourist spots Random trip suggestions User limited maximum days in a journey 41

42 Contact information Homing Tam Department of Land Surveying and Geo Informatics The Hong Kong Polytechnic University Q&A Session 42