Querying and Managing Complex Data

Abstract

Database technology has advanced to support increasingly complex data -- from relations to semi-structured data and unstructured documents. More recently, graph databases have regained attention following demands from applications like social networks and recommendation systems. Graph analysis, usually associated with the Complex Networks field, has become a central tool in areas such as biology, physics and linguistics. Database systems should improve support to these data and applications beyond the data model level tackled by current graph databases, including more flexible querying models and management mechanisms. In this proposal, we define the characteristics of the highly interconnected data that underlies many of these modern applications. We adopt the term complex data as a reference to the field of complex networks. Our objective is to propose querying and management mechanisms for this type of data. A data management system for complex data requires a flexible query model that explores the topology of the relationships, taking into account their eventual uncertainty. Efficient query processing becomes a challenge, requiring new mechanisms for relationship-based query optimizations. The system also requires a relationship-centric architecture, with support to management of the life-cycle of the relationships. Here we show several aspects of our ongoing research aimed at meeting the new requirements. Our solution models complex data as property graphs with weighted relationships. We propose a new query language that allows ranking of elements based on properties of the topology of the graph. The queries are evaluated based on a variation of the spreading activation model, which is the core of the query processor and the main target for query optimization strategies. To simplify data management and expand query expressiveness, we introduce mappers, which are responsible to encapsulate the logic of relationship creation in our framework. The research has produced 4 papers that describe experiments with real data, demonstrating the practicability of our approach.