We present a tunable two-dimensional class hierarchy indexing technique (2D-CHI) for object-oriented databases. We use a two-dimensional file organization as the index structure. 2D-CHI deals with the problem of clustering objects in a two-dimensional domain space consisting of the key attribute domain and the class identifier domain. In conventional class indexing techniques using one-dimensional index structures such as the B/sup +/-tree, the clustering property is exclusively owned by one attribute. These indexing techniques do not efficiently handle the queries that address both the attribute keys and the class identifiers. 2D-CHI enhances query performance by adjusting the degree of clustering between the key value domain and the class identifier domain based on the precollected usage pattern. For performance evaluation, we first compare 2D-CHI with the conventional class indexing techniques using an analytic cost model based on the assumption of uniform object distribution, and then verify the cost model through experiments using the multilevel grid file as the two-dimensional index. We further perform experiments with nonuniform object distributions. Our experiments show that our proposed method does indeed build optimal class index structures regardless of query types and object distributions. We strongly believe that our paper significantly contributes to building a self-tunable database system by supporting automatically tunable index structure.

It hsa been known that different representations of a query ret rieve different sets of documents.Recent work suggests that significant improvements in retrieval performance can be achieved by combining multiple representations of an information need.However, little effort has been made to understand the reason why combining multiple sources of evidence improves retrieval effectiveness.In this paper we analyze why improvements can be achieved with evidence combination, and investigate how evidence should be combined.We describe a rationale for multiple evidence combination, and propose a combining method whose properties coincide wit h the rationale.We also investigate e the effect of using rank instead of similarity on retrieval effectiveness.