Research On A Data Model For Complex Relationships And Context-Dependent Information

Our work is from the project Theory Foundation and System Implementation of Unstructured Data Management funded by National Natural Science Funds for Foreign Distinguished Young Scholar (Grant No.60688201). The thesis mainly focuses on its core theory part such as designing the data model to provide explicit modeling and characterization of all kinds of relationships between meta data of unstructured data and to efficiently store, manage, and maintain unstructured data, and query language to search unstructured data using these relationships. We summarize our work and contributions as follows:Based on the systematic investigation of information modelling requirement and related works, we have proposed a novel powerful data model called Information Net-working Model (INM) which allows us to represent various kinds of relationships be-tween objects, between objects and relationships, and between relationships naturally and directly, and support context-dependent access to object properties. The main novel feature of INM is the introduction of role relationships and novel mechanisms to represent complex relationships between objects and the context-dependent information to reflect the dynamic, many-faceted, and evolutional aspects of real-world entities in a natural and direct way. A role relationship has two functions. On one hand, it is a relationship to connect objects in two classes. As a relationship, it covers all the char-acteristics in traditional data models. It can also form relationship hierarchies, specify inverse relationship, and automatically generate context in terms of the inverse. On the other hand, it is also a role that the objects in one class play in objects in another class. As a role, it is different from the role in role models in which a role is represented by the membership relationship between instances and role classes. Thus, it can only express simple semantics. In INM, a role can get more rich and elaborate context by specifying the role label and nesting them into the context obtained from the inverse above. More-over, a role relationship can induce a role relationship class with the same name, which denotes a set of instances that participate in the role relationship. Also, the context-dependent information can be automatically generated by specifying context-dependent properties and nesting them into the context of the corresponding role relationship. An-other novel feature of INM is that all the information regarding a real-world entity is grouped in one object with a unique object identifier instead of scattering in a hierarchy of objects as in role models. Also, the object can belong to several classes to reflect the dynamic, many-faceted, and evolutional aspects of the entity in a natural and direct way and can have context-dependent properties nested within the object. Thus it provides one-to-one correspondence between the real world and the corresponding information model.Also, we have designed a concise, compact, and expressive modeling language for INM. It provides powerful constructs to express the rich semantics and integrity con-straints associated with various relationships. In schema, we use two kinds of relation-ship specifications to denote four kinds of relationships, their inverse, and relationship hierarchies in a compact style. In instance, we use nested tuples and relationship hierar-chies to construct complex relationships and context-dependent information of objects. Moreover, we present the formal semantics of modeling language in following four as-pects. Firstly, the relationship specification decomposition discusses how a specification is decomposed into a set of or a pair of definitions/specifications that are related to a rela-tionship and its inverse. Secondly, the inheritance contains role relationship inheritance at the schema and instance level, object class inheritance, and role relationship class inheritance. Thirdly, the relationship inverse demonstrates the semantics of inverse tak-ing inheritance with overriding into account, which is used in relationship consistency constraints. Finally, constraints are used to guarantee an intended semantics of the database. The inheritance hierarchy should not exist cycles. Also, each object assign-ment in the database must be well-typed with respect to corresponding class definitions. Moreover, each relationship assignment must be consistent in terms of inverses. Partic-ularly, our modeling language supports built-in semantics for consistency between two objects participating in a relationship. None of existing languages support this feature and thus require tedious and manual operations to maintain the consistency.Based on INM, we have also designed a powerful, concise, compact, comprehensible, and declarative query language called IQL which has a uniform style with INM modeling language and can search the information in a simple, direct, and natural way. It strictly separates the query part from the result part, directly uses INM object structure to represent queries, and allows variables to match anything in this structure. It can not only explore and traverse the natural graph structure of objects to extract and construct meaningful results but also avoid cycles by utilizing the specific semantics of objects. Also, the use of logical variables and the mechanism for not distinguishing attributes and all kinds of relationships etc., make IQL flexible and easy to use in practice. Moreover, on the premise of maintaining the same expressive power, built-in semantics of exist and negation makes the syntax of IQL more simple to use.We have also developed a database management prototype system for INM, which enable users to add, delete, and update the schema and instance with ISL (Information Specification Language) and IML (Information Manipulation Language) that are based on our modeling language, and to query all kinds of information with IQL.