Re(c)SPARQL: Recommendation Engine SPARQL

The flexibility of the Resource Description Framework (RDF) allows us to model any kind of knowledge. Being able to produce recommendations from RDF data sources by taking advantage of semantic knowledge has several advantages. For instance a user could be assisted in selecting information from the Web, reducing the information overload.

Current approaches for producing recommendations typically consist of separating the process of data retrieval and computation of recommendations. The latter task is achieved by a fixed component which uses the retrieved data as input. The result is a recommender which is specialized for specific domains like music, books, movies, etc. but which cannot be easily customized or reused and which is therefore not able to handle the diversity and unstructuredness of Semantic Web data.

In this project we aim to produce recommendations from RDF-graphs by extending SPARQL, the official query language for RDF-graphs. SPARQL has been designed to retrieve explicit information from RDF-graphs in contrast to recommendations or suggestions which are inherently implicit information. This novel approach allows us to write highly parameterizable queries. The current recommendation model allows us to compute recommendations using classic approaches such as content-based, collaborative filtering or hybrid.

The implementation of an RDF recommender repository based on Sesame (RecSesame), which allows us to evaluate Re(c)SPARQL queries, has demonstrated the feasibility of our approach. We are currently investigating how to improve the recommendation model in order to allow for new and more flexible queries.

Re(c)SPARQL: Recommendation Engine SPARQL

Downloads

Project Members:

References