FUTURE TRENDS

The management of incomplete information in multidimensional databases is a young field with more challenges than solutions. The most important open challenge is perhaps implementation. Many of the techniques outlined in this chapter appear to be easy to implement at low cost, and will add value to a multidimensional database product. But implementation in this area lags behind the research. Some research prototypes are available. In particular, the solutions we outlined for handling incomplete metadata (i.e., partial classification hierarchies) can alleviate a frustrating problem for many users (Pedersen et al., 2000).

Multidimensional databases are complex systems, and introducing a technique to handle incomplete information will impact other areas. For example, solving the exclusive disjunctive value problem by adding hidden units changes the storage size estimation for eager implementations (Deshpande et al., 1997). For semi-eager multidimensional databases, allowing imprecise values in the hierarchy means that new strategies must be devised to determine which points in the multidimensional space should be materialized, i.e., should incomplete values be materialized (Yang et al., 1997; Baralis et al., 1997)? Another area of open research is in the design of visualization tools to judge the amount of incomplete information in a multidimensional database and to direct queries to (more) complete areas. Finally, multidimensional databases have evolved with a limited set of aggregates, drawn from aggregates in relational databases. In contrast, the field of statistics is rife with techniques for aggregating and reasoning with incomplete data. Borrowing more heavily from the field of statistics would enhance multidimensional database research and practice.