Despite the rapidly increasing availability of open data as Linked Data on the Web, a lot of data is still published in other, non-RDF formats, such as HTML forms/tables, or via Rest APIs. To use this data as linked data, one would have to transform it into RDF triples and store them in a triple store every time a source gets updated. The rst OBDA/ RDB2RDF systems started to emerge recently, providing a solution for creating virtual RDF graphs on top of relational data using ontologies and mappings, such as Ontop [1], Ultrawrap [2] and Sparqlify1. In this work, we introduce a framework for extending existing OBDA techniques to support querying of data from di erent sources that are available on the Web, such as webtables and Rest APIs.
mappingId webtable_rotten_tomatoes target rot:{rank} rot:rank {rank}; rot:title {Title} . source select rid as rank, Title from
webtable('http://www.rottentomatoes.com/ top/bestofrt/', 3) The mapping described above encodes how the attributes of the resulting virtual table (i.e., title and rank in our example) are mapped into RDF terms. In this way, users are able to pose queries that retrieve the rank and the title of rotten tomatoes movies by querying the respective HTML table directly, using SPARQL. select distinct ?title where { ?s rot:title ?title . ?s2 rot:rank ?rank }
When a SPARQL query as the one described above is posed, it gets translated into the respective SQL query that contains the virtual table operator that corresponds to the data source (i.e., webtable). The operator populates the virtual table with data and after the whole SQL query is evaluated the results are returned as virtual RDF terms to the users.
To improve performance, we allow the option to use cached data for a time window w. If w is expired, fresh data needs to be retrieved. The length of w is a parameter that can be con gured per mapping (i.e, di erent mappings can access the same data source at di erent rates depending on the needs).
Ontop4theWeb is based on the system Ontop2 [
We have implemented virtual operators that access Twitter and Foursquare APIs, and we have extended the MadIS webtable operator that retrieves data from HTML tables.
To evaluate our approach, we conducted experiments that involved SPARQL queries that access di erent sources (Twitter API, Webtables) as well as generated datasets (i.e, generated webtables to measure scalability) and we measured the query execution time. Even in cases where no cache is used, Ontop4theWeb can execute queries in a few seconds, whereas when the cache is used, queries are executed in less than a second. The experiments also showed that Ontop4theWeb is able to query webtables with up to 100,000 rows within minutes. 2 https://github.com/ontop/ontop 3 http://ontop-spatial.di.uoa.gr/ 4 http://madgik.github.io/madis 5 http://www.sqlite.org [2] Sequeda, J.F., Miranker, D.P.: Ultrawrap: SPARQL execution on relational data. J. Web Sem. 22 (2013) 19{39 [3] Bereta, K., Koubarakis, M.: Ontop of Geospatial Databases. In: Proceedings of the 15th International Semantic Web Conference. (2016) [4] Chronis, Y., Foufoulas, Y., Nikolopoulos, V., et al: A Relational Approach to Complex Data ows. In: EDBT/ICDT Workshops. (2016)