Towards full end-users control of social recommendations
1. Towards full end-users control
of social recommendations
* Presenter
gbosetti@lifia.info.unlp.edu.ar
Gabriela Bosetti *, Sergio Firmenich, Alejandro Fernández, Martin Wischenbart,
Gustavo Rossi & Damiano Distante
LIFIA, Facultad de Informática, UNLP - Argentina
Josephinum Research - Austria
Unitelma Sapienza - Italy
3. Recommender Systems
Recommender systems (RS) are a kind of information filtering system
that seeks to present the information that are likely to interest the
user.
Social RS ultimately helps users to cope with the challenges of the
social overload.
3
4. Web-based RS
➔ Sites not offering recommendations
➔ Sites offering recommendations
◆ that are homogeneous
◆ from the same source
◆ that can’t be fully customized
● in certain pages of the site
● when recommendations are available
● what is recommended
● considering a concrete group of users
4
5. Controllability in RS
Giving the user –with domain knowledge– explicit control
over the weights of the items and their friends increases the
quality of recommendations and the user satisfaction [3].
Users prefer
recommendations
received after they had a
certain level of control [5].
A substantial portion of users
choose to switch among
recommendation algorithms
until they find the one which
satisfies them the most [6].
5
8. An approach for...
Allowing end-users to collaboratively generate and use a
recommendation layer above the presentation layer of any
existing Web application, giving more control on the
recommendations creation and retrieval.
The approach rests on two pillars:
1. Any Web content can be used as recommendation content
2. Any Web site can be augmented with recommendations
8
9. The approach
1. Users are empowered to annotate arbitrary Web content
as domain-specific objects employing semantic tagging.
Resulting objects from various websites become the
source of recommendations.
9
10. The approach
Recommended
2. Any website can be augmented with a recommendation
layer supported by a configurable recommendation service,
and providing heterogeneous recommendations.
10
12. A support toolset
➔ 3 webextensions + backend
➔ In-situ, in any Web site
➔ 100% end-user oriented
Define a
template
Extract an
info item
Retrieve
info items
as rec.
12
27. What’s next?
1. An experiment with end users to assess its usability and
potential of adoption.
2. Extend the toolset
a. Automatically complete information items’ data when
something is missing on a Web page
b. More recommendation algorithms to be used from the
configuration options by the widgets
c. Automate the identification of templates
27
28. G. Bosetti *, S. Firmenich, A. Fernández, M. Wischenbart, G. Rossi & D. Distante
Thanks!
http://tiny.cc/icwe2018
29. References
Reference numbers used in this presentation are consistent with
those used in the paper:
Bosetti, G., Firmenich, S., Fernández, A., Wischenbart, M., Rossi, G., &
Distante, D. (2018, June). Towards Full End-Users Control of
Social Recommendations. In International Conference on Web
Engineering (pp. 304-311). Springer, Cham.
https://link.springer.com/chapter/10.1007/978-3-319-91662-0_24
Extra slides
https://docs.google.com/presentation/d/177HarJ4CXHhSnViX_3bQEchK
2Dji3VLq99oc1SAoWXY/edit?usp=sharing
29