Machine Learning-Based Query Augmentation for SPARQL Endpoints

Disclaimer

This site contains the data associated to the paper entitled Machine Learning-Based Query Augmentation for SPARQL Endpoints, published at the WEBIST 2018 conference. This paper as well as the materials included in this site are original and have been not publised before. You are not allowed to keep any dataset(s). You must wait till we make public the dataset(s) with a DOI under a Creative Commons license.

Abstract

Linked Data repositories are becoming an increasingly popular source of publicly-available data. However, users accessing these data through SPARQL endpoints usually face significant response times. A commonly used solution to improve access times to data repositories is query augmentation, which is modifying queries to retrieve more data that is potentially relevant to subsequent requests. In this paper we propose a novel approach to query augmentation for SPARQL endpoints based on machine learning. Our approach separates the structure of the query from its contents and measures two types of similarity, which are then used to predict the structure and contents of the augmented query. We test our approach with real-world query logs from the Spanish and English DBpedia and the results indicate that the proposed approach yields high-accuracy prediction. We also show that, by caching the predicted queries, we can retrieve data relevant to several subsequent queries at once, achieving a higher cache hit rate than previous approaches.

Data

In this section we include the files containing the data used in our experimental setup.

Query Type (q-type) Prediction

The dataset for this experiment is found in the file querytypes.7z (2.2MB). This file contains the Weka data files (.arff) used to predict the query type of the next query using different numbers of previous queries. Each file contains a feature vector consisting of the query types of previous queries (called t1, t2, ...) and the time interval between the queries (i1, i2, ...).

Triple Patterns Prediction

The dataset for this experiment is found in the files triplepatterns_5prev.7z (2.8MB) and triplepatterns_10prev.7z (4.7MB). The folders contain the Weka data files (.arff) used to predict the triple patterns of the next query using 5 and 10 previous queries respectively. Each file represents a different triple pattern prediction (1st, 2nd, ...) and contains a feature vector consisting of the triple patterns of previous queries (t1, t2, ...) and the time interval between queries.

Reproductibility: This is our study in openml.org

Code

There is no source code. All has been done using Weka. Have a look at our study in openml.org.

About the authors

Mariano Rico	(mariano.rico@upm.es) Ontology Engineering Group, UPM
Rizkallah Touma	(rizk.touma@bsc.es) Supercomputing Center (BSC), Barcelona, Spain
Anna Queralt	(anna.queralt@bsc.es) Supercomputing Center (BSC), Barcelona, Spain
María S. Pérez	(mperez@upm.es) Ontology Engineering Group, UPM

Acknowledgements

This work has been supported by the European Union's Horizon 2020 research and innovation program (grant H2020-MSCA-ITN-2014-642963).