Approximate matchings in scientific databases

Organizations often need access to scientific data stored in independently managed databases. In this paper, we analyze the data heterogeneity problem which occurs when the data conveying the same or similar information is represented differently in different databases. We introduce the matching join to process queries in scientific databases and discuss the three steps to evaluate it. First we transform the query using the functional dependencies in the database to incorporate additional knowledge. Second, we use rules and weights to compare the attributes. Matching joins can also be used to obtain approximate answers. In the third step, we propose a numeric measure, called the comparison value, c, to estimate the quality of matching and suggest deterministic and probabilistic ways of deriving it. Finally, we analyze the problem of estimating the cutoff value for c that would minimize the cost of errors during the join computation.<>