Searching for related objects in relational databases

To discover knowledge or retrieve information from a relational database, a user often needs to find objects re- lated to certain source objects. There are two main chal- lenges in building an efiective object search system: The huge amount of objects in the database and the large num- ber of difierent relationships between objects. In this paper we introduce Ross, an eficient and accurate relational object search system. Ross accepts complex queries that enable users to specify the relationships among objects. To measure the relationships of join paths, Ross consid- ers the difierent semantics of difierent joins, and com- bines both selectivity and lengths of join paths to measure their strengths. A novel approach is used to find the best join paths between relations, which converts the database schema into a graph, so that the shortest paths in the graph correspond to best join paths in the database. Ross uses a stream-based system architecture to handle com- plex queries containing logical operators, which can find the most related objects upon users' requests. Compre- hensive experiments are conducted to show the high scal- ability and efiectiveness of Ross.