Knowledge via Graph Reasoning

Abstract: The idea of the present project is to research novel algorithms for interacting with graph databases. In particular, we aim at implementing a next generation search and recommendation engine on top of a graph representation. For decades, researchers and companies have tried to accommodate connected, semi structured data into relational databases. But whereas relational databases are perfectly suited to model rigid and tabular structures, they fail when attempting to model the ad hoc, exceptional relationships that dynamically occur in the real world. Actually, graph databases can overcome this and other major limitations. The first focus is the development of algorithmic procedures that build an interface from a graph database to diverse data repositories covering disjoint and overlapping domains (big data with respect to variety/volume). This interface will automatically carry out both extractions of information and subsequent translations into nodes and edges of the graph. The second building block is the development of a novel paradigm for searching and browsing knowledge. As our data representation is based on a native graph, we are able to implement several innovative features. First, by means of explicit/implicit relationships in the graph model, we can answer non-trivial queries and find relevant connections in the first place. Second, graphs offer a natural way of representing rich content, which in turn increases the possibility for exploratory interactions. Third, topological descriptors allow dynamic rankings of nodes and empower us to find relevant data with high precision. Finally, the schema free nature of graph databases allows for dynamic evolvements of the data basis.