Introduction

Currently the amount of data has increased excessively with a variety of formats. The Semantic Web technology aims to provide standards and facilitate analyzing such big data. The Semantic Web uses RDF data to describe the data on the web in form of Subject-Predicate-Object called "triples" [1] that makes the data to have the standard data model.

In the present, there are many approaches to store and query RDF data. One approach to store RDF data is Triple Store designed for storing the triples format of RDF data [2] and queried by using SPARQL query language. However, from the Berlin Benchmark results [3], Triple Stores show poor performance when compared to the relational database systems. NoSQL database removes some features of relational databases and uses other data models to improve the performance of database. This has motivated many works to store RDF data by using NoSQL database.

This paper reviews existing systems and designs a framework to store RDF data in NoSQL database. One of the main goals is to design a Semantic Web application framework that uses RDF data with NoSQL database, i.e., MongoDB. The ultimate objective is to provide a better support for researchers in developing the Semantic Web applications.

Review of NoSQL-based RDF Database

This section reviews some of RDF database systems that use NoSQL to store the RDF data including Neo4j [4] , AllegroGraph [5] , H2RDF [6] , Oracle NoSQL [7] , MonetDB [8] and CumulusRDF [9]. The comparison is based on some criteria of database software such as Implementation language, Database Model, SPARQL1.0, SPARQL1.

Framework Design

This section describes our design for an application framework representing system architecture that compares the Triple Store-based implementation with the NoSQLbased implementation. We also provide query translation that represents some example translation of basic SPARQL queries adapted from the Berlin Benchmark [3] to MongoDB queries.

In a system architecture based on the OAM framework [10], we compare between Triple Store based implementation and NoSQL based implementation. The Triple store based implementation uses Jena TDB to store the RDF data and OAM API that uses SPARQL to query the data from Jena TDB. In NoSQL based implementation, we use RDF to JSON-LD Converter to convert RDF data format to JSON-LD format, which is JSON-based format designed for Linked data [11], and use JSON-LD Parser to parse and import JSON-LD data to MongoDB. The OAM API then uses MongoDB query API to query the data from MongoDB.

Table 2 illustrates some query translation based on the Berlin SPARQL benchmark. In Table 2, query 1 shows an example of query using FILTER, ORDER and LIMIT. Query 2 shows an example of query using OPTIONAL. Query 3 shows an example of query using regular expression. This paper has proposed the design of RDF database system by using MongoDB to store the data in JSON-LD format and its query API. In the future, we will conduct the performance comparison of Triple Store, MongoDB RDF Database, and relational database using the Berlin SPARQL Benchmark. Several techniques will be investigated to improve the performance of the MongoDB RDF Database.