1. ComfRide - A Smartphone based Comfortable Public Route Recommendation
Authors: Rohit Verma, Surjya Ghosh, Mahankali Saketh, Niloy Ganguly, Bivas Mitra, Sandip Chakraborty
Indian Institute of Technology Kharagpur, India

2. Understanding Commuter Comfort
Different people have different comfort preferences which is seen all over the world.
Countries
Delay in Reaching
No seat
Bad Road
Traffic Jam Re O R VB B M N
India
40.7
43.5
15.8
47.8
37.2 15
24.9
42.3 30
2.8
67.6
26.5
5.9
Nepal 80 20 40 60
Iran
12.5
62.5 25 75
37.5
Sri
66.7
16.7
16.6
83.3
33.3 50
USA
42.9
57.1
14.2
14.3
28.6
France
Survey reveals importance of different features for commuters’ comfort

3. But Is this a Major Problem?
More than 60% of the source, destination pairs have at least 4 routes between them
Approximately 25% have more than 8 routes
Cumulative Distribution of number of bus routes between random (source, destination) pairs calculated
For every number of routes there is considerable variation in trip distance
For none of number of routes, the mean value of standard deviation exceeds 15 minutes

4. Objective
Develop an end-to-end smartphone based personalized bus route recommender system, which recommends the most comfortable bus route based on commuters’ comfort choices.

5. Data Collection: Road Information
Speed Breakers
Turns
Bus Stops
Etc.

6. Data Collection: Route Information
0.6
0.68
0.5
0.7
0.37
0.62
0.57
Congestion
Jerkiness of Bus
Probability of getting seat
Etc.

7. Annotated City Map
0.6
0.5
0.7
0.3
0.8

8. Selecting the best route
Source
0.6
0.5
0.7
0.3
0.8

9. Selecting the best route
Source
0.6
0.5
0.7
0.3
0.8
Destination

10. Selecting the best route
Source
Comfort Parameters
Probability of sitting
Waiting time at bus stop
0.6
0.5
0.7
0.3
0.8
Destination

11. Recommend the best Route!
Selecting the best route
Source
Comfort Parameters
Probability of sitting
Congestion
0.6
0.5
0.7
0.3
Recommend the best Route!
0.8
Destination

12. Selecting the best route
2.1
0.6
0.5
6.3
7.8
1.6
0.6
0.4
0.5
0.6
1.6
0.6
0.5
1.7
0.4
0.5
0.6
0.5
0.3
0.5
1.1
1.3
1.7
0.4
0.6
1.6
0.6
0.4
1.4

13. A Major Challenge
Such systems, e.g. FAVOUR[1] and PaRE[2], have very high memory requirement and computation time.
This is not permissible for mobile application
[1] Paolo Campigotto, Christian Rudloff, Maximilian Leodolter, and Dietmar Bauer Personalized and situation-aware multimodal route recommendations: the FAVOUR algorithm. IEEE Transactions on Intelligent Transportation Systems 18, 1 (2017), 92–102.
[2] Yaguang Li, Han Su, Ugur Demiryurek, Bolong Zheng, Tieke He, and Cyrus Shahabi PaRE: A System for Personalized Route Guidance. In Proceedings of the 26th International Conference on World Wide Web. 637–646

14. A Major Challenge Solution: Using DIOA
Utilize Dynamic Input Output Automata (DIOA) which prunes and updates the graph dynamically based on the context of a query.
DIOA has a set of internal actions which dynamically includes only those nodes and edges of the route graph which are required as per the query.
Such systems have very high memory requirement and computation time.
This is not permissible for mobile application

15. DIOA Advantages: Explained
1. State space reduction:
For any graph based approach the state space would be;
S = T * NC2 * FCf
where, T is no. of time zones, N is number of stops, F is total number of features and f is the features used.
But for SIOA, the state space would be;
S = 1 * NC2 * f

16. FAVOUR: 3 x 10 2 x 8 5 B4 T1 B1 B1, B4 B1, B4, B2 B2 B2, B3 B1, B4, B2

17. FAVOUR: 3 x 10 2 x 8 5 PaRE: 3 x 10 x 8 5 B4 T1 B1 B1, B4 B1, B4, B2

18. FAVOUR: 3 x 10 2 x 8 5 PaRE: 3 x 10 x 8 5 ComfRide: 1 x 10 x 5 B4 B1T1 B1
B1, B4
B1, B4, B2 B2
B2, B3
B1, B4, B2 B2 T2 B3 B3 B3 B3

19. FAVOUR: 3 x x 8 5
PaRE: 3 x 10 x 8 5
ComfRide: 1 x 10 x 5

20. System Architecture Feature Ordering
User Input
Feature Ordering
Optimized graph generation by pruning unwanted nodes
Recommended Route

21. Working of the DIOA

22. Break Journey: A Special Scenario
We utilize Dynamic Programming approach.
Let C = 𝑠𝑒𝑡 𝑜𝑓 𝑎𝑙𝑙 𝑝𝑜𝑠𝑠𝑖𝑏𝑙𝑒 𝑏𝑟𝑒𝑎𝑘𝑝𝑜𝑖𝑛𝑡𝑠
Then a DAG is G(V,E) is constructed as follows;
𝑉=𝐶 {𝑠, 𝑑} and
𝐸= 𝑒 𝑣1 ∈𝑉, 𝑣2 ∈𝑉 𝑒 𝑣1 ∈𝑉, 𝑣2 ∈𝑉 𝑡ℎ𝑒𝑟𝑒 𝑖𝑠 𝑎 𝑏𝑢𝑠 𝑟𝑜𝑢𝑡𝑒 𝑓𝑟𝑜𝑚 𝑣1 𝑡𝑜 𝑣2 }
The optimal substructure for n breaks is thus given as
𝜒 𝑠, 𝑑 𝑛 = max 𝑐 𝜖 𝐶, 1 ≤𝑚 ≤𝑛 1 𝑛 𝑛−𝑚 × 𝜒 𝑠, 𝑐 𝑛−𝑚 +𝑚 × 𝜒 𝑐, 𝑑 𝑚
So, if a commuter prefers β breaks, then the maximum RCI is given as;
𝜒 𝑠, 𝑑 𝑚𝑎𝑥 = max 0 ≤𝑛 ≤ 𝛽 𝜒 𝑠, 𝑑 𝑛
Consider the breakpoints during a travel

23. Durgapur
Kolkata
Bhubaneswar
ODISHA
Experiment setup
50 volunteers were recruited for data collection across three cities, Kolkata, Bhubaneswar, Durgapur
Some volunteers were given specific routes while others followed their regular routes
Every trip was taken by a group of volunteers on different days at various time of the day.
They travelled through both the ComfRide recommended route as well as the Google Maps (G-Maps) recommended route (least expected time).

24. Evaluation - Deployed at Kolkata, 4 S-D Pairs
Source
Destination
Distance (km)
Average Travel Time (min) 1 2 3 P1 KM RH 10 15 55 65 P2 SC 6 30 P3 Gh 8 - 45 50 P4 JP 70

25. Evaluation: Contd.. Personalized Features (PaRe)
Sitting Probability
Number of breakers
Number of bus stops
ComfRide recommended routes differ from G-Maps recommended routes for many instances, which is as high as 70% for P2 and P3
FAVOUR gives priority to the general choices of the commuters over a route, and so, fails to capture the personal choices of a commuter
PaRE gives priority to the personal choices, and thus ignores environmental impacts
Congestion
Jerkiness of road
General Features (FAVOUR)

26. Conclusion
The key concept behind ComfRide is to embed the general awareness and intelligence used by a regular commuter to choose the best (comfortable) bus route to reach her desired destination.
ComfRide recommended routes have on average 30% better comfort level than Google navigation based recommended routes.

27. Thank you!
ComfRide:
Follow the work of Complex Network Research Group (CNeRG), IIT KGP at:
Web:
Facebook:
Twitter: