软件工程外文文献—嵌套SQL的能力.docx

外文原文 1. Introduction In Imielinski and Mannila (1996) we have argued that due to the tremendous number of association rules which are generated by data mining there is a pressing need to provide rule querying, both to selectively generate rules from data as well as query the rulebase of rules which were previously generated. Several such query languages have been proposed in the literature (Han et al., 1996; Meo et al., 1996). Other similar efforts were also recently noticed in literature, either to generate rules satisfying some constraints in the database (Ng et al., 1998) or to integrate mining closely with relational databases (Sarawagi et al., 1998). In our view, although these proposals are a step in the right direction, they still lack a few features such as closure, using the full expressive power of SQL, and treating rule querying in a manner similar to rule generation. Following is the set of primitives which, in our opinion, a rule query language should satisfy: ? Ability to nest SQL: Since SQL is a well accepted interface to relational databases, and allows for declarative, set level, expressions which are optimized substantially, it is desirable if users can utilize SQL primitives such as sorting, group-by, and others like these within MSQL, and be able to express nested queries using the SQL nested query constructs like [NOT] IN, [NOT] EXISTS etc. ? Closure: Mining is essentially an iterative task, which often involves refifinement of existing queries and regenerating results. The language must provide for operations to further manipulate the results of the previous MSQL query. ? Cross-over between data and rules: To permit the “iterative refifinement” as mentioned in the previous item, the language must allow primitives which can map generated rules back to the source data, and vice-versa. ? Generation versus querying: Given the enormous size of rulebases, it may not always be possible to “extensionally” maintain rulebases. The language should a