Cross-Domain Recommender Systems

Cross-Domain Recommender Systems

The goal of a Recommender System (RS) is to provide useful suggestions for a user and therefore to support her in a decision-making process. Whereas the vast majority of approaches have been designed to generate recommendations of items that belong to a single domain, online stores like Amazon or eBay or even social networking services are capable of collecting a large amount of user preferences for different domains, e.g. books, clothes, music, electronics, etc. It would be therefore more than beneficial to leverage this information and to deliver recommendations that span across several domains. There is an increasing interest in Cross-Domain Recommender Systems (CDRS) and new underlying models are currently being studied.

In this regard, graph models might be useful to transfer the knowledge from one domain to the other, especially if items in different domains are in some way connected. This is the case for Dbpedia, a project aiming at extracting structured content from Wikipedia's articles. This dataset is available as an RDF-Graph.

Master Project

The goal of the project is to implement, compare and evaluate different approaches to produce cross-domain recommendations.

Datasets

For the master project we will use data collected from Facebook about personal interests in three domains: movies, books and music. The original data was crawled with the purpose of organizing a challenge at the ESWC 2015 conference. We created an RDF dataset from it that contains incoming and outgoing Dbpedia edges for those items and resources connected by means of up to three predicates.

Curriculum

Master of Science: 3rd Semester (Teamproject / Masterproject)
ECTS: 16