7/27/2019 Poster Flarty Acmmm2013
1/1
Flarty:
recommending art routes using check-ins latent topics
Alberto Del Bimbo, Andrea Ferracani, Daniele Pezzatini
Media Integration and Communication Center, University of Florence, Italy
larty framework
ACM Multimedia 2013, Barcelona, October 21-25, 2013 demo session
MICC Media Integration and Communication Center / [email protected], [email protected], [email protected], [email protected]
arty is a mobile location-based social network for the dynamic
onstruction and recommendation ofart routes in cities, via item
ased similarity algorithms, places topic extraction and user
nterest modeling.
arty enables users to follow the routes in realtime and toomplete the individual steps through check-ins. The user prole is
ferred from check-ins.
ach user is described mainly by two vectors: a vector of topics
xtracted by the places he checked-in and a vector of the ratings
xpressed on these check-ins.
To deal with the so called cold start scenario issue the system
provides some initial gamication techniques to have a rst usermodel: at the very rst access the user is asked to give a rating from
1 to 5 on some representative POIs.
Cold Start Scenario
arty proposes to users two types of art routes:custom: calculated weighting results of recommendation
gorithms and clustering techniques according to user interests, and
classic: predened routes of POIs must see.
he custom routes are created starting from the results returned by
he Recommender module that are further elaborated and ltered
y the results of the Clustering module.
he users anity with the set of POIs suggested by the
ecommender is rened comparing their representative topics with
he topics extracted from users check-ins. Then routes are generated
onsidering geo-location and proximity. Each POI is represented in
he model by its own Wikipedia page. The algorithm used for
rouping is CVB, Collapsed Variational Bayesian inference
gorithm for LDA.
The application provides three dierent routes visualization
paradigms: a list view, a map view and an augmented reality view.
The POIs that are part of the routes can be ltered and customized
manually through addition/deletion/reordering.
Several ltering criteria can be applied (e.g. distance and averagevote) to allow the user to optimize the routes on the basis of their
preferences.
Every time the user selects a route he is continuously
localized through GPS and guided by the application in the visit.
Views
larty