Huge volumes of invaluable information are hidden behind web relational
databases. They could not be extracted by search engines. The problem is
especially severe for long text data, for example: book reviews, company descriptions,
and product specifications. Many researches have investigated to integrate
information retrieval and database indexing technologies to provide keyword
search functionality for these useful contents. Due to diversifying data relationships
in application domains and miscellaneous personal preferences, current
ranking results of related researches do not satisfy user requirements. We design
and implement a Weight-Adjustable Ranking for Keyword Search (WARKS)
system to address the issue. Mean average precision (MAP) and mean rank reciprocal
difference (MRRD) are proposed as measurements of ranking effectiveness.
We use an integrated international trade show database as our experimental
domain. User study demonstrates that WARKS performs better than previous
practices.
