SeRela: Exploiting Semantic Relationships for Cross-Domain Recommendations
In this project we investigate methods that leverage the semantic relationships between items in addition to other aspects considered by traditional RSs to produce cross-domain recommendations on top of RDF-graphs.
With the goal of generating recommendations that span across multiple domains, some cross-domain approaches
require a model in which items of a different nature as well as the semantic relationships between
those items' attributes are represented. The flexibility of the Resource Description Framework (RDF)
allows one to model this kind of knowledge while keeping a semantic meaning in the representation, which is both, machine-processable and human-readable.
In addition to this, hundreds of RDF-based datasets of different domains have been published according to Linked Open Data (LoD) principles thus reducing the need of having to build such knowledge-model required by Recommender Systems (RS) from scratch. Whereas some researchers have already envisioned Linked Data as a solution to the problem of acquisition, aggregation and linkage of knowledge, most of them focus on the existence of semantic relationships across domains leaving aside the semantics encoded in them.
In this project the problems range from that of finding complementary products to point-of-interest recommendation. Several challenges have to be addressed such as the mapping of items or attributes to RDF resources or the interconnection sparsity occurring in practice in LoD datasets which affects the quality of recommendations.
Current Bachelor (BT) / Master Theses (MT) / Projects (P)
- (MT) Combining point of interest (POI) recommendation with real-time route planning (Ongoing)
- (P) Cross-domain recommendations in RecSesame (Ongoing)
Finished Master Theses
- Leveraging Linked Open Data to find complementary Products (Completed on August 2016)
- Cross-domain Recommendations using Dbpedia (Completed on March 2016)
Victor Anthony Arrascue Ayala, Anas Alzoghbi, Martin Przyjaciel-Zablocki, Alexander Schätzle, Georg Lausen:
Speeding up Collaborative Filtering with Parametrized Preprocessing [ .pdf ]
Proc. of the 6th International Workshop on Social Recommender Systems (SRS 2015),
in conjunction with the 2015 ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2015). Sydney, Australia, August 2015.
Victor Anthony Arrascue Ayala, Martin Przyjaciel-Zablocki, Thomas Hornung, Alexander Schätzle, Georg Lausen:
Extending SPARQL for Recommendations [ .pdf ]
Proc. of the 6th International Workshop on Semantic Web Information Management (SWIM 2014),
in conjunction with the 2014 ACM International Conference on Management of Data (SIGMOD 2014). Snowbird, Utah (USA), June 2014.