Adaptive Query Relaxation and Result Categorization Based on Data Distribution and Query Context

To address the empty and/or many answer problem of Web database query, this paper proposes a general framework to enable automatically query relaxation and result categorization. The framework consists of two processing parts. The first is query relaxation. In this part, each specified attribute is assigned a weight by measuring the query value distribution in the database. The rarely distribution of the query value of the attribute indicates the attribute may important for the user. The original query is then rewritten as a relaxed query by expanding each specified attribute according to its weight. The second part is result categorization. In this step, we first speculate how much the user cares about all attributes (including specified and unspecified attributes) under the query context by using the KL-divergence. Then, the categorizing attribute in each level of the tree can be determined according to its importance for the user. The most important attribute should be the categorizing attribute for the first level of the navigational tree. Lastly, the navigational tree is generated automatically and presented to the user so that the user can easily select the relevant tuples matching his/her needs. Experimental results demonstrated that the query relaxation method can achieve the Precision of 78% and 75% for UsedCarDB (used car dataset) and HouseDB (real estate dataset), respectively, and the result categorization method can also achieve the lowest total and averaged navigational costs than the existing categorization methods.