Example of data modeled in RDF format 

Contexts in source publication

Context 1

... of web data, while SW data have an important density of valuable information that can be used for enriching business analysis (Thi et Nguyen, 2008 ; Kä mpgen et Harth, 2011 ; Zorrilla et al. , 2012 ; Etcheverry et R. A. Vaisman, 2012 ; Abelló et al. , 2013 ; Ibragimov et al. , 2014 ; Aufaure et Chiky, 2014). Combining BI with SW, however, is not a trivial task due to the scalability, complexity and heterogeneity of SW data. It raises the following questions: How to integrate heterogeneous SW data in a BI system originally designed for factual data? How to carry out multidimensional analyses over large amount of SW data in the lack of relevant model? How to present analysis results containing both factual data and SW data? These questions are examples of issues waited to be resolved. The aim of this paper is to present an up-to-date survey of research results and outline future research challenges in BI and SW domains. The rest of the paper is organized as follows. We (i) briefly present the concepts of BI and SW in the section 2; (ii) give an overview of recent research results combining the domain of BI with SW in the sections 3 and 4; (iii) discuss emerging trends and perspectives of future researches in the section 5. The term of Business Intelligence (BI) refers to a set of techniques used for collecting, extracting and analyzing business data to support decision-making process. Coming from heterogeneous and distributed operational sources, data used in decision-making process are stored in Data Warehouse after going through a process called ETL (standing for Extraction, Transformation and Loading). Among different types of data warehouse, On-Line Analytical Processing (OLAP) data warehouse has been a specific research topic for over a decade. The concepts of OLAP were firstly proposed in (Codd, Codd, et Salley, 1993), they provide solutions for creating, managing, analyzing and reporting large amount of multidimensional data in an interactive way. Among all data models proposed for OLAP, the Star Schema (Kimball, 1996) is the most widely accepted model (Chaudhuri, Dayal, et Narasayya, 2011). At conceptual level, Star Schema presents data according to subjects of analysis (facts) and axes of analysis (dimensions). At logical level, Star Schema can be built on top of different types of databases: Multidimensional OLAP (MOLAP), Relational OLAP (ROLAP) and Hybrid OLAP (HOLAP). At physical level, Star Schema can be implemented in different ways, as long as the implementation conforms to the twelve evaluation rules defined in (Codd, Codd, et Salley, 1993), such as multidimensionality, transparency, accessibility, etc. Together with the multidimensional data model, a set of operators is indispensable for OLAP analysis. They permit to aggregate information (Drilldown, Rollup), filter analysis results (Slice, Dice) and change analysis axes (Pivot). (Kimball, 1998) points out that the main advantages of OLAP model lie in its simplicity and understandability that permit users to interact with large amount of complex data in an efficient way. Nowadays, OLAP is a well-mastered technology when it comes to homogenous and structured data in classical data warehouse. However, as factual data provide only limited and partial views over open-world business scenarios (Zorrilla et al. , 2012), the data warehouse community looks for solutions for enriching data collection with external data. To accurately exploit web data, a system needs to be capable to read the exact semantic meaning of web-published information. An acknowledged way to publish machine-readable information is to use Semantic web (SW) technologies. The purpose of SW technologies is to fix a common vocabulary and a set of interpretation constraints (inferring rules) so as to semantically express metadata over web information and allow doing some reasoning on it. These technologies 2 provide the capability of annotating web data with semantics, e.g., through RDF and ontologies, hence generating a web of semantic linked data (e.g., Linked Open 3 Data cloud ). 4 Tim Berners-Lee pointed out four principles that SW data should follow : use Uniform Resource Identifiers (URIs) to identify object; use Hypertext Transfer Protocol (HTTP) to facilitate searching for objects by human-beings; use the 5 Resource Description Framework (RDF) format as standard to provide descriptive information about an object; link URIs to others in order to connect individual data into a data web. Compared to traditional web technologies which focus mainly on data representation, SW puts a higher value on providing machine-readable information about web resources and relationships between resources. More specifically, SW presents human knowledge through structured collections of information and sets of inference rules (Berners-Lee, Hendler, et Lassila, 2001). The basic data model is RDF permitting to express simple statements about resources, using named properties and values (cf. figure 2). Resources described by RDF are not necessarily retrievable on the web, they can be anything with an unique identity, from physical objects to abstract concepts (McBride, 2004). A Triple Store permits to store RDF data. The set of statements in a RDF Triple Store is composed of URIs, blank nodes and literals. A RDF triple refers to subject , predicate and object : a subject is a web resource identified by a URI or a blank node; an object can be a web resource or a literal that possesses a primitive value; a predicate is a binary relationship connecting a subject with an object. For instance, in the figure 2 we can find the predicate denoted by the label Concerns associating the resource Sales with another resource ProductX , and another predicate named hasPrice connecting the subject denoted ProductX to a textual literal “ 30 ” which is the product ’ s price. There exist other SW formats with more powerful expressivity than RDF. Built 6 on top of RDF, RDF Vocabulary Description Language (or RDF schema or RDFS ) is a language that defines the terms used in RDF graph. Equivalent to schema definition language in relational and object-oriented data model, RDFS is used to describe classes of resources. In other words, RDFS is a simple ontology definition language which allows expressing taxonomies. The concepts of RDFS are described in form of a set of predefined RDF resources with special meanings. However, the reasoning capacity of RDFS is very limited, only basic inferences about taxonomies are supported (Horrocks, Patel-Schneider, et van Harmelen, 2003). Facing to this issue, the Web Ontology Working Group of W3C develops more powerful ontology languages, such as OWL-Lite, OWL-DL, OWL-Full, which allows defining explicit, formal conceptualizations of domain models. In general, OWL enhances the expressivity of RDF and RDFS schema by adding Description Logic (DL). Hence, OWL is an ontology language with sufficient expressive power which can support efficient reasoning through well-defined syntax and semantics (Antoniou et van Harmelen, 2004). By using the SW formats, web resources can be enriched with annotations and other markups capturing the semantic metadata of resources. However, not all current technologies are fully compatible with the semantic enrichment. For instance, traditional Information Retrieval (IR) technologies cannot directly exploit the annotated semantic meaning of web resources (Finin et al. , 2005). On the other hand, new research directions have been proposed to combine traditional research approaches with SW technologies, such as Semantic Information Retrieval (Ferná ndez et al. , 2011), Exploratory OLAP (Abelló et al. , 2015) etc. In this paper, we only focus on the emerging research direction which aims at enhancing traditional BI with new SW technologies. Nowadays, a large number of researches try to merge OLAP analysis with SW technologies both in data integration and data processing levels. This research direction permits to combine powerful tools and technologies in both domains. But it is not a trivial work mainly due to the reason that follows: OLAP requires a specialized data model to support multidimensional analysis over aggregated values of measurements at different granularity levels. However, SW does not dispose of appropriate model fully satisfying criteria about hierarchical levels proposed by (Codd, Codd, et Salley, 1993). Carrying out OLAP analysis directly over SW data is difficult and inefficient by the lack of suitable data model bridging the gap between SW and OLAP domains. Actually, OLAP is originally conceived for analysis over homogenous and stable warehoused data. With arrival of profusion of schema-less Web information, data become more and more heterogeneous and volatile. By mentioning the volatility of SW data we refer to the quick, unceasing and unpredictable changes in SW data sources. Traditional OLAP technologies are challenged while being applied to analyses over SW data. Facing to these issues, lots of research efforts have been made to combining OLAP with SW. Two types of approaches can be identified (Figure 3). The first approach is OLAP-analyses oriented, which consists of extracting, transforming and then storing multidimensional SW information in traditional OLAP data warehouses (§ 3.1), so that it can be analyzed through existing OLAP tools. The second approach is multidimensional modeling oriented, whose aim is to carry out OLAP analyses directly over RDF-like data modeled in an appropriate multidimensional format (§ 3.2). At the end of the section, we provide a conclusive table (cf. Table 1) that summarizes all mentioned work. OLAP analyses are carried out through analysis operators, such as roll-up , drilldown , rotate and so on (Ravat et al. , 2008). Analysis results are usually presented in Multidimensional Table (MT) allowing visualizing several analysis axes around a subject. Based on a MT, decision-makers can further carry ...

Context 2

... of SW data in the lack of relevant model? How to present analysis results containing both factual data and SW data? These questions are examples of issues waited to be resolved. The aim of this paper is to present an up-to-date survey of research results and outline future research challenges in BI and SW domains. The rest of the paper is organized as follows. We (i) briefly present the concepts of BI and SW in the section 2; (ii) give an overview of recent research results combining the domain of BI with SW in the sections 3 and 4; (iii) discuss emerging trends and perspectives of future researches in the section 5. The term of Business Intelligence (BI) refers to a set of techniques used for collecting, extracting and analyzing business data to support decision-making process. Coming from heterogeneous and distributed operational sources, data used in decision-making process are stored in Data Warehouse after going through a process called ETL (standing for Extraction, Transformation and Loading). Among different types of data warehouse, On-Line Analytical Processing (OLAP) data warehouse has been a specific research topic for over a decade. The concepts of OLAP were firstly proposed in (Codd, Codd, et Salley, 1993), they provide solutions for creating, managing, analyzing and reporting large amount of multidimensional data in an interactive way. Among all data models proposed for OLAP, the Star Schema (Kimball, 1996) is the most widely accepted model (Chaudhuri, Dayal, et Narasayya, 2011). At conceptual level, Star Schema presents data according to subjects of analysis (facts) and axes of analysis (dimensions). At logical level, Star Schema can be built on top of different types of databases: Multidimensional OLAP (MOLAP), Relational OLAP (ROLAP) and Hybrid OLAP (HOLAP). At physical level, Star Schema can be implemented in different ways, as long as the implementation conforms to the twelve evaluation rules defined in (Codd, Codd, et Salley, 1993), such as multidimensionality, transparency, accessibility, etc. Together with the multidimensional data model, a set of operators is indispensable for OLAP analysis. They permit to aggregate information (Drilldown, Rollup), filter analysis results (Slice, Dice) and change analysis axes (Pivot). (Kimball, 1998) points out that the main advantages of OLAP model lie in its simplicity and understandability that permit users to interact with large amount of complex data in an efficient way. Nowadays, OLAP is a well-mastered technology when it comes to homogenous and structured data in classical data warehouse. However, as factual data provide only limited and partial views over open-world business scenarios (Zorrilla et al. , 2012), the data warehouse community looks for solutions for enriching data collection with external data. To accurately exploit web data, a system needs to be capable to read the exact semantic meaning of web-published information. An acknowledged way to publish machine-readable information is to use Semantic web (SW) technologies. The purpose of SW technologies is to fix a common vocabulary and a set of interpretation constraints (inferring rules) so as to semantically express metadata over web information and allow doing some reasoning on it. These technologies 2 provide the capability of annotating web data with semantics, e.g., through RDF and ontologies, hence generating a web of semantic linked data (e.g., Linked Open 3 Data cloud ). 4 Tim Berners-Lee pointed out four principles that SW data should follow : use Uniform Resource Identifiers (URIs) to identify object; use Hypertext Transfer Protocol (HTTP) to facilitate searching for objects by human-beings; use the 5 Resource Description Framework (RDF) format as standard to provide descriptive information about an object; link URIs to others in order to connect individual data into a data web. Compared to traditional web technologies which focus mainly on data representation, SW puts a higher value on providing machine-readable information about web resources and relationships between resources. More specifically, SW presents human knowledge through structured collections of information and sets of inference rules (Berners-Lee, Hendler, et Lassila, 2001). The basic data model is RDF permitting to express simple statements about resources, using named properties and values (cf. figure 2). Resources described by RDF are not necessarily retrievable on the web, they can be anything with an unique identity, from physical objects to abstract concepts (McBride, 2004). A Triple Store permits to store RDF data. The set of statements in a RDF Triple Store is composed of URIs, blank nodes and literals. A RDF triple refers to subject , predicate and object : a subject is a web resource identified by a URI or a blank node; an object can be a web resource or a literal that possesses a primitive value; a predicate is a binary relationship connecting a subject with an object. For instance, in the figure 2 we can find the predicate denoted by the label Concerns associating the resource Sales with another resource ProductX , and another predicate named hasPrice connecting the subject denoted ProductX to a textual literal “ 30 ” which is the product ’ s price. There exist other SW formats with more powerful expressivity than RDF. Built 6 on top of RDF, RDF Vocabulary Description Language (or RDF schema or RDFS ) is a language that defines the terms used in RDF graph. Equivalent to schema definition language in relational and object-oriented data model, RDFS is used to describe classes of resources. In other words, RDFS is a simple ontology definition language which allows expressing taxonomies. The concepts of RDFS are described in form of a set of predefined RDF resources with special meanings. However, the reasoning capacity of RDFS is very limited, only basic inferences about taxonomies are supported (Horrocks, Patel-Schneider, et van Harmelen, 2003). Facing to this issue, the Web Ontology Working Group of W3C develops more powerful ontology languages, such as OWL-Lite, OWL-DL, OWL-Full, which allows defining explicit, formal conceptualizations of domain models. In general, OWL enhances the expressivity of RDF and RDFS schema by adding Description Logic (DL). Hence, OWL is an ontology language with sufficient expressive power which can support efficient reasoning through well-defined syntax and semantics (Antoniou et van Harmelen, 2004). By using the SW formats, web resources can be enriched with annotations and other markups capturing the semantic metadata of resources. However, not all current technologies are fully compatible with the semantic enrichment. For instance, traditional Information Retrieval (IR) technologies cannot directly exploit the annotated semantic meaning of web resources (Finin et al. , 2005). On the other hand, new research directions have been proposed to combine traditional research approaches with SW technologies, such as Semantic Information Retrieval (Ferná ndez et al. , 2011), Exploratory OLAP (Abelló et al. , 2015) etc. In this paper, we only focus on the emerging research direction which aims at enhancing traditional BI with new SW technologies. Nowadays, a large number of researches try to merge OLAP analysis with SW technologies both in data integration and data processing levels. This research direction permits to combine powerful tools and technologies in both domains. But it is not a trivial work mainly due to the reason that follows: OLAP requires a specialized data model to support multidimensional analysis over aggregated values of measurements at different granularity levels. However, SW does not dispose of appropriate model fully satisfying criteria about hierarchical levels proposed by (Codd, Codd, et Salley, 1993). Carrying out OLAP analysis directly over SW data is difficult and inefficient by the lack of suitable data model bridging the gap between SW and OLAP domains. Actually, OLAP is originally conceived for analysis over homogenous and stable warehoused data. With arrival of profusion of schema-less Web information, data become more and more heterogeneous and volatile. By mentioning the volatility of SW data we refer to the quick, unceasing and unpredictable changes in SW data sources. Traditional OLAP technologies are challenged while being applied to analyses over SW data. Facing to these issues, lots of research efforts have been made to combining OLAP with SW. Two types of approaches can be identified (Figure 3). The first approach is OLAP-analyses oriented, which consists of extracting, transforming and then storing multidimensional SW information in traditional OLAP data warehouses (§ 3.1), so that it can be analyzed through existing OLAP tools. The second approach is multidimensional modeling oriented, whose aim is to carry out OLAP analyses directly over RDF-like data modeled in an appropriate multidimensional format (§ 3.2). At the end of the section, we provide a conclusive table (cf. Table 1) that summarizes all mentioned work. OLAP analyses are carried out through analysis operators, such as roll-up , drilldown , rotate and so on (Ravat et al. , 2008). Analysis results are usually presented in Multidimensional Table (MT) allowing visualizing several analysis axes around a subject. Based on a MT, decision-makers can further carry out OLAP operators to continue their analyses. OLAP operators are only applicable to specialized data structures (Harinarayan, Rajaraman, et Ullman, 1996 ; Ravat et al. , 2008 ; Etcheverry et R. A. Vaisman, 2012), RDF descriptions, however, do not dispose component that can directly support OLAP analysis. For instance, in order to carry out drilldown and rollup operations, we need to represent data according to hierarchical levels within a dimension. However, even though RDF triple can be used to describe web resources and relationships between them (instance level), it does not allow revealing hierarchical ...