Close and Loose Associations in Keyword Search from Structural Data

Keyword search over structural data enables users to seek information from databases without knowing the structure of data or mastering actual query languages like SQL. In a keyword query, data items or text attributes are matched to the keywords and the result of a query is typically a set of graphs consisting of connected tuples. The result should be ranked which means that the text attributes and connections must be scored and combined. Typically, the length of a connection is the main criterion in ranking the connections, i.e. shorter connections are scored higher than longer ones. The length of a connection is usually based on the foreign key references but their direction has received less attention. At the conceptual level, cardinality constrains correspond to foreign key references or their combination. In the present paper, we investigate the effect of the combinations of cardinality constrains on the result of a keyword search. We find that the combination of cardinality constraints indicates how close the association between keywords is. We also show that the Minimal Total Joining Network of Tuples (MTJNT) principle loses semantic connections or fragments the results of a keyword search from relational databases.