Recommender System use-cases

Personalized recommender systems take user profiles into account when the prediction for particular user and item is generated. The prediction techniques for recommender systems can be divided into three main categories: content-based, collaborative-based and hybrid prediction techniques, which combine the content-based and collaborative-based methods. Content-based predictions analyze interactions between particular user and all the items in the system. These recommendation techniques use knowledge of the user's past experience to recommend further items the user might be interested in. Collaborative filtering predictions analyze interactions between all users and all items. They analyze user's past behavior and by analyzing the past behavior of other users, they find users with similar characteristics and recommend items.

Here we demonstrate two recommender systems use-cases:

  • A Recommender System for the platform
  • Generic web-service recommender system template for RapidAnalytics

A Recommender System for the platform

The VideoLectures challenge was organized as Discovery challenge of  the ECML-PKDD 2011 conference.

The challenge has been organized over specialized data mining competitions portal and was running from April to July 2011. Over 300 hundred teams registered, and more than 60 were active participants. A workshop organized around solutions of winners and participants of the challenge was held in September at the ECML-PKDD 2011 conference in Athens. The workshop proceedings are available online.

The aims of the challenge were multifold: from impact on recommender systems  research community, to improvement of the current recommender system of the VL.Net site, and as a main testbed for which soultions in terms of computational workflows were sought within e-LICO the platform.

Some of the solutions from the winning submission of the challenge have been built into an improved recommender system of the VideoLectures.Net site. The development platform implementing winning recommender at VideoLectures site is available at:

A recommender system built into the VideoLectures.Net portal

The dataset of the challenge, together with task and evaluation tools is publicly available from the

Generic web-service recommender system template for RapidAnalytics

Feneric web-service recommender system arhitecture

Recommendation extension for RapidMiner enables construction of diverse recommendation workflows within Rapid Miner. However, the basic application of these workflows should be within recommendation engines or web services. Rapid Analytics execution engine provides the platform for the realization of web services based on Rapid Miner workflows.

Figure above depicts a generic solution based on Rapid Analytics and recommendation workflows. Two recommender workflows (ONLINE and OFFLINE) are based on identical recommender operators but differ only in the learning option. They change the contents of the relational recommendation SQL database and refresh recommendation model in a mutually coordinated fashion. Front-end web-service is responsible for fast retrieval and recommendation to web-page/users.