Different data sources may be designed differently. Moreover, the
design of the same data set may evolve over time. If the results of a keyword query or natural language interface
depend on some specific styles of design, it will perform poorly for the data sets that do not follow those styles of
design. Hence, no matter how the data is designed, a keyword query interface should return the same answers as long as
the data set contains the same information. We introduced and formally defined this property of design independence
for keyword query interfaces. Design independence provides a metric to measure the degree
of logical data independence sought by the architects of modern database models.
For a given set of content, the ranking produced by a design independent query
interface is invariant under equivalence-preserving reorganizations of the structure of the data. We also analyzed this
property for current keyword query interfaces and proved that they do not generally provide design independence.
We then went on to develop a novel query interface that is design independent. In addition to providing theoretical
results for the worst case, we conducted an extensive empirical study to measure the design independence property
of current and the newly proposed method using real-world data sets. We showed that our new method has higher
average design independence than other methods. We also performed an extensive user study and showed that the
new query interfaces deliver the same or better ranking than other methods.