Advanced

Recommender System Validation Platform

Ullén, Johan LU (2015) In LU-CS-EX 2015-25 EDA920 20151
Department of Computer Science
Abstract
With most applications where recommender systems are used, it is impor-
tant that they produce a better result than a system with no recommender, or
one with a previous recommender. Deploying an untested system, even to a
smaller user sample can be very costly if the system produces negative results.
It is often in a developer’s interest to create several candidate systems. They
need some way of comparing recommender systems before selecting one or
a few to launch. While the methods of testing have been explored, and their
statistical soundness motivated, in other work, it is not obvious
how to do it in practice. This report describes the implementation of a modular
and configurable framework, and analyses this framework with two... (More)
With most applications where recommender systems are used, it is impor-
tant that they produce a better result than a system with no recommender, or
one with a previous recommender. Deploying an untested system, even to a
smaller user sample can be very costly if the system produces negative results.
It is often in a developer’s interest to create several candidate systems. They
need some way of comparing recommender systems before selecting one or
a few to launch. While the methods of testing have been explored, and their
statistical soundness motivated, in other work, it is not obvious
how to do it in practice. This report describes the implementation of a modular
and configurable framework, and analyses this framework with two different
cases. The experimentation shows the power of how such a framework can
be utilized to reduce overhead work when approaching evaluation of a new
recommender system. (Less)
Please use this url to cite or link to this publication:
author
Ullén, Johan LU
supervisor
organization
course
EDA920 20151
year
type
H3 - Professional qualifications (4 Years - )
subject
keywords
generic framework, validation, evaluation, recommender system
publication/series
LU-CS-EX 2015-25
report number
LU-CS-EX 2015-25
ISSN
1650-2884
language
English
id
7445203
date added to LUP
2015-06-25 07:55:43
date last changed
2015-06-25 07:55:43
@misc{7445203,
  abstract     = {With most applications where recommender systems are used, it is impor-
tant that they produce a better result than a system with no recommender, or
one with a previous recommender. Deploying an untested system, even to a
smaller user sample can be very costly if the system produces negative results.
It is often in a developer’s interest to create several candidate systems. They
need some way of comparing recommender systems before selecting one or
a few to launch. While the methods of testing have been explored, and their
statistical soundness motivated, in other work, it is not obvious
how to do it in practice. This report describes the implementation of a modular
and configurable framework, and analyses this framework with two different
cases. The experimentation shows the power of how such a framework can
be utilized to reduce overhead work when approaching evaluation of a new
recommender system.},
  author       = {Ullén, Johan},
  issn         = {1650-2884},
  keyword      = {generic framework,validation,evaluation,recommender system},
  language     = {eng},
  note         = {Student Paper},
  series       = {LU-CS-EX 2015-25},
  title        = {Recommender System Validation Platform},
  year         = {2015},
}