Representation Independent Proximity and Similarity Search

Finding similar or strongly related entities in a graph database is a fundamental problem in data management and analytics with applications in similarity query processing, entity resolution, and pattern matching. Similarity search algorithms usually leverage the structural properties of the data graph to quantify the degree of similarity or relevance between entities. Nevertheless, the same information can be represented in many different structures and the structural properties observed over particular representations do not necessarily hold for alternative structures. Thus, these algorithms are effective on some representations and ineffective on others. We postulate that a similarity search algorithm should return essentially the same answers over different databases that represent the same information. We formally define the property of representation independence for similarity search algorithms as their robustness against transformations that modify the structure of databases and preserve their information content. We formalize two widespread groups of such transformations called {\it relationship reorganizing} and {\it entity rearranging} transformations. We show that current similarity search algorithms are not representation independent under these transformations and propose an algorithm called {\bf R-PathSim}, which is provably robust under these transformations. We perform an extensive empirical study on the representation independence of current similarity search algorithms under relationship reorganizing and entity rearranging transformations. Our empirical results suggest that current similarity search algorithms except for R-PathSim are highly sensitive to the data representation. These results also indicate that R-PathSim is as effective or more effective than other similarity search algorithms.