Based on the vast domain resources of RDF(S) on the web and SPARQL's powerful query ability, this article presents a new method of designment of E-R model. The steps for this design are: (1) Formulating SPARQL rules (including resource query rules and schema query rules) by the analysis of RDF(S)'s structure. (2) Parsing the optimal resource obtained through the query sentences. (3) Completing the designment by taking advantages of the translation from RDF(S) model to entity-relationship model in accordance with the content queried. The results indicate that, the designment of E-R model based on RDF(S) could restore user real requirements of great possibilities and help database designer to complete design in a strange area.
Generally speaking, database design depends, to a large extent, on the designer's understanding and representing of user requirements[1]. Passed through 60 years development, the main ideas about database construction have improved a lot. Much research has been put forward. Kahn raised a way of obtaining description information in the process of database design[2], which emphasized the designer should gather requirement in the real world and judge the requirement's su ciency and completeness during the process of design. Blaha, Premerlani came up with the method of Object-Oriented to build relational database[3], it promoted adherence to normal forms and improved integration between databases and applications. Finkelstein, Schkolnick took advantage of the calculation method of heuristic to optimize data dictionary and applied it in relationship database design[4]. Asuman, Birol designed a generalized expert system for database design[5], they optimized the expert system by adding new design approach and modifying older method and formed the human interaction mode. Unfortunately, because of the shock of the web, the traditional methods can not longer t the needs in some certain mode, so it's confronted with more complicated problems.
Resource Description Framework (RDF)[6] provides a set of data models to support the description of domain resources stored on the web and has strong ability on the knowledge sharing. SPARQL[7] just o ers the powerful query capabilities to RDF(S) and could be able to analyze RDF deeply. Based on vast amounts of RDF resources and SPARQL, this article brings a fresh perspective about designment of E-R model. As a result, the designment further strengthens the database design system and satis es the user in special elds a lot. 2
Based on the strong ability on knowledge sharing, we choose RDF as design
source. Particularly, we use WordNet[8] and Swoogle[9] to obtain user
requirements. In this article, we propose a designment of E-R model based on RDF(S)
and it is built around the traditional method. There are three main tasks:
(
For task (
For the designment of E-R model by RDF(S), we parse RDF resources by SPARQL to transform the elements from RDF(S) model to E-R model. As SPARQL mainly uses the rules in WHERE, so the core content of this article is focus on the query rules. To keep things simple, the rules below mentioned keep only the part of WHERE, the parts of PREFIX, SELECT and FROM are all ignored. 3.1
Like RDF, SPARQL also provides the matching pattern of triple. Unlikely, the triple's subject, predicate and object all can be variable. The triples in SPARQL are used to match the triples in RDF and there will be the results if they matches successfully. Because every RDF le has its own namespace and builds the le structure under it, so we can not query all of the information with a set of inherent rules. The designer has to design the query sentences by the speci c namespace. In this case, we put forward the resource query rules in table 1.
In order to have a comprehensive understanding of the le, we can get all the triples by rule 1. As needed, we may often run into the situation that there are one known data item and two unknown or two known data items and one unknown, so we design rule 2 and rule 3. According to the content above, we can make a traverse only by one known data item. Based on the diversity of the semantic, we can add some constraints and deformation to the rules to make the semantic more richer. In rule 4, we use UNION to combine two triples together. As a result, there will be at least one branch matching the triples in RDF. In rule 5, we use OPTIONAL to combine two triples together. Speci cally, the second triple will modify the rst. In rule 6, we use FILTER to combine two triples together and the second triple will constrain the rst. Furthermore, the constraints can be logical expressions of Boolean value.
Specially, there will be the blank nodes in the results. The blank node is a special variable and it only matches the element whose data type is blank. Otherwise, the blank node doesn't represent the real meaning and it only holds up some space to connect the other two nodes. During the process of query, we will meet the blank nodes inevitably. At this time, designer should ignore the blank node and skip it to search the bilateral nodes. 3.2
About RDF, in addition to the user de ned namespace, RDF Schema also provides its own namespace for the layout of the structure of RDF. The rules mentioned above describe the resource rules, but the results are always disorderly and they are even meaningless in extreme cases. On the other hand, resource query is always blind and has low e ciency. Therefore, we put forward the schema query rules shown in table 2. So, we design the query sentence that one known data item connects with one unknown by the prede ned vocabulary RDF Schema provides. Thus, as long as the designer understand the special semantics of the prede ned vocabularies, he can obtain the unknown data item from the known easily. On the whole, the schema query rules not only improve the readability of RDF but also enhance the regularity of the query. Specially, like resource query rules, we can also add some constraints and deformation to schema query rules to complete more complex query and they are no longer to be described. Bene ting by the hierarchical structure, we simulate the whole RDF le as the tree model and regard the triples as the nodes of the tree. For the elements we have found without redundancy and missing, this article takes the strategy of top-down and the method of gradually precision to query. Furthermore, Width First Traversal method is also used during the whole process. It is necessary to state that with the query nodes expanding, the semantic relevance between the follow-up nodes and the initial nodes will decrease rapidly. So, in order to pledge the usefulness, the designer should de ne the layer of query with actual requirement to improve the quality of query results. 4.1
In order to achieve the model-to-model transformation, it's necessary for us to
analyze the models' structures with their characters to ensure the relation of the
speci c elements between the two models. As the source, RDF has rich semantic
content and can describe the domain knowledge in more detail. We will get
rdfs:class which represents class and rdf:Property which represents property by the
generalization of rdfs:Resource. About the RDF(S) model, the most important
characteristic is hierarchy, the class can be inherited by rdfs:subClassOf, while
the property by rdfs:subPropertyOf. Using the tree structure, rdfs:Resource acts
as the root node while rdfs:Class and rdf:Property and their subclasses act as
the sub-nodes to expand. Because RDF is a kind of resource to store
information, so we save the classes and the attributes after they have been instantiated.
For stipulating the class connected with property and the relationship between
class and property, RDF provides rdfs:domain and rdfs:range to constrain the
property's domain and range respectively. Based on the constraints, we will be
able to nd the relationship between the class and the property as well as the
relationship between the classes. Finally, the RDF network will be presented
clearly. The structure is showed in Figure 2.
(
When traversing, the classes act as the starting points and the ones searched in the rst case stored in queues. Then, putting out the rst class and nding its sub-nodes as well as the sibling-nodes until the sub-nodes are all searched. After the traverse of the rst class, we put out the second one and do the same process. Likewise, making a traverse for the whole tree. While, because of the namespace user de ned, we may not get classes by rule 1 directly sometimes. So the designer should traverse from the result's rst record. By the instance and the property mentioned in the rst record, we will nd the related class.
(
Due to the semantic absence of query, some properties only connects with instances rather than classes. In this case, we can only query by instances to nd the relationship between property and class.
(
(
Due to the truth that RDF is an open semantic framework, writers are not under the speci c discipline to construct classes, properties and instances. Thus, they could use the namespace they have de ned or RDFS prede ned. In the reason that there is little or not any RDFS's prede ned vocabularies in the RDF les, designer often needs resource query by the requirements. This leads to the situation that some elements are unlikely to be acquired. On the contrary, ER model makes demands on the form and structure of data with uni ed and re ne requirements. Base on the standard, all the elements are designed strictly to follow the regulation and they are not permitted by default. Therefore, it is inevitable to cause semantics absence when doing the transformation. 5 5.1
As design result, relational database has the ability of data storage and o ers a standard to manage resources. Entity, relationship and attribute together constitute the E-R model whose structure is shown in Figure 3. The relationship de nes the relation between entities and it extends the large network structure by connecting entities. An entity has several attributes, while the attributes modify the entity and express the entity's features. Among those, when an attribute or a set of attributes can uniquely determine the entity, we call it primary key. Specially, the relationship between the entities has many types, such as one-toone (1:1), one-to-many (1:n) and many-to-many (m:n). The types express the correspondence of the entities. 5.2
Model-to-model transformation
(
(
(
Note:VARCHAR stores the variable length strings with the maximum length of 255 characters. If the length is greater than 255 characters, we use TEXT.
In the actual process, the query and the transformation is a process of the two at the same time, there is no obvious sequence. In combination with the resource query rules and the schema query rules, we parse RDF of the tree structure and facilitate the transformation from RDF(S) model to E-R model. Figure 4 shows the corresponding relationship between RDF(S) model and E-R model.
The article chooses the inference engine Jena[?] to assist SPARQL with querying and reasoning. The RDF le and the SPARQL le are used as input and the query results are used as output. Designer selects the content that he wants according to the RDF resources and writes some corresponding query sentences into SPARQL le.
As a typical example, we choose "blogger" whose address is "http:// wiki.creativecommons.org/ Special:ExportRDF/ Blogger" given by IBM to act as source le to design. Speci cly, we use resource query rules and schema query rules to parse the le and the layer of query is de ned at 5. Based on rule 1, we can list all of the triples which is shown in Figure 5. However, according to rule 7, we can not nd any classes, so we traverse from the result's rst record.
As we can see, the rst record gives a description to the instance of "<http:// blog.planetrdf.com/rss.xml>". Therefore, we use rule 9 to query the instances type and the result is "rss:channel". Unfortunately, we don't have found some properties connected with "rss:channel". For this reason, we turn to give a query to the instance of "<http://blog.planetrdf.com/rss.xml>" again. With rule 3, we can get the property ("foaf:topic") of "rss:channel". In addition, its value and data type which is "string". Then, we transform it to "VARCHAR" which is the data type of SQL. What's more , we can also nd the relationship "foaf:maker" that connects "rss:channel" and a blank node. What is shown in Figure 6. According to the query sentence like f < http://blog.planetrdf.com/rss.xml > foaf:maker :b0.
:b0 ?y ?z. g, we will get the resource "<http://blog.planetrdf.com/>" connected with "rss:channel" by the blank node. Otherwise, we can also search for "foaf:Agent" which represents the blank node and its properties "foaf:name" and "foaf:interest". By convention, we regard "foaf:name" as primary key and underline it. What is shown in Figure 7. Based on the query about the classes "foaf:maker" connects with, we can ensure the cardinality is "m:n". What is shown in Figure 8.
Repeating the steps, we draw the E-R model about "blogger" in Figure 9. 7
Designment of E-R model based on RDF(S) is a new method for database design. It takes advantages of RDF's ability on knowledge sharing and makes use of the resources stored on the web vastly. The results show that the method could restore user real requirements of great possibilities and help database designer to complete design in a strange area.