Integrating Travel Planning and OnTour Support in a CaseBased Recommender System

1
Integrating Travel Planning and On-Tour
Supportin a Case-Based Recommender System

• Francesco Ricci, Dario Cavada and Quang Nhat
  Nguyen
• eCommerce and TourismResearch Laboratory
• ITC-irst
• http//ectrl.itc.it

Partially supported by Fondazione Cassa di
Risparmio di Trento e Rovereto
2
Content

• Application Scenario
• Intelligent Travel Recommender pre travel
  component
• Objectives of mITR mobile extension
• Usage Scenario of the mobile component
• Query refining based on user feed-back
• Technological infrastructure
• Conclusions

3
Application Scenario

• Pre-travel ITR
• User leisure traveler
• Task select a destination for a residential
  holiday in Trentino plus an accommodation and
  additional attractions.
• Output a travel bag containing a set of travel
  products (booked or simply chosen)
• On-Tour mITR
• When the traveler is On-Tour he/she is interested
  in some complementary services that fits well
  into the plan. In this case the details like
  distance, cost or opening time are very
  important.

4
ITR (Intelligent Travel Recommender)
General Wishes
Add to travel bag
5
Objectives of mITR

• The ITR cannot be simply adapted to a mobile
  system.
• Develop methodologies and software components for
  a mobile application that can extend and
  integrate ITR methodologies to support on-tour
  travelers.
• Methods for interactively refining a user request
  exploiting explicit user feedback given to sample
  products shown to him and (partially) satisfying
  the initial request.
• Reduce the time and the cognitive effort required
  to successfully complete a recommendation session.

6
Usage Scenario
ITR
The Traveler Creates a Travel Plan
Add the Restaurant to the case base
Get Restaurants(userID, location, time)
Requests restaurants
mITR
Shows best candidates
Updating query and re-ranking
Feed-back (critic) one feature
The new ranked list is shown
Stores result to the local bag
Selects a restaurant
mStorage
7
Query update cycle
Queryexecution
Logical and similarity query
Result list(ranked items)
User selectone?
Item added to plan
Feedback and query update
New logical and similarity query
8
User feedback
Feedback from the user
Result list (ranked list)
Features list of the item selected
9
Initial set of candidates
The system performs the logical query (CityTN)
and retrieves the initial set of candidates from
the restaurant catalog. The rank is given by
computing the similarity of the items in the
result set with those contained in past similar
sessions.
Recommender System DB
Result Set
QL(CityTN)
logical query
10
Query update
Update the initial query using feed-back from the
user. Initial query (city TN) (?,?,?)

• If the user select the
• I like thai cuisine then(city TN) (? , ? ,
  Thai)
• I would like a less expensive one(city TN)
  (cost 20 - d) (? , ? , ?)
• I would accept a more expensive one(city TN)
  (cost 20 d) (? , ? , ?)

11
Query update example
(CityTN) (?,?,?)
d(x,y) di(x1,y1) dn (xn,yn)
if feature is unknown d 1 If nominal feature
and xi ? yi d 1 If nominal feature and xi
yi d 0 If Numerical feature (xn-yn/range)
di(xi,yi)
Metric used for similarity ranking
(CityTN) (?,?,Thai)
12
Related work

• L. M. Ginty and B. Smyth. Deep dialogue vs casual
  conversation in recommender systems. In
  Recommendation and Personalization in eCommerce,
  Proceedings of the AH2002 Workshop, pages 8089,
  Malaga, Spain, May, 28th 2002. Universidad de
  Malaga.User selects one item among many
  simultaneously shown then the selected item is
  used for query updating (not applicable on small
  screen devices).
• H. Shimazu. Expertclerk Navigating shoppers
  buying process with the combination of asking and
  proposing. In B. N. (Ed.), editor, Proceedings of
  the Seventeenth International Joint Conference on
  Artificial Intelligence, IJCAI 2001, pages
  14431448, Seattle, Washington, USA, August 4-10
  2001.User gives feedback to the whole product
  and not to the single feature. It may will take
  long time to converge to a good solution.
• R. Burke. Knowledge-based recommender systems. In
  J. E. Daily, A. Kent, and H. Lancour, editors,
  Encyclopedia of Library and Information Science,
  volume 69. Marcel Dekker, 2000.Very similar
  query refining methods Not a mobile application
  and not using long term preferences (past cases)

13
Infrastructure
14
Conclusions
We have shown an integrated recommender system
(pre-travel and on-tour)

• It uses the ITR system for retrieving an initial
  set of best candidates.
• It adopts a combination of logical and similarity
  query for ranking the result.
• It tries to minimize the interaction length using
  the feed-back from the user for query refining.

15
Thanks
Research partially supported by Fondazione
Cassa di Risparmio di Trento e Rovereto