Estimating Recall and Precision for Vague Queries in Databases

Abstract

Abstract.In vague queries, a user enters a value that represents some real world object and expects as the result the set of database values that represent this real world object even with not exact matching. The problem appears in databases that collect data from different sources or databases were different users enter data directly. Query engines usually rely on the use of some type of similarity metric to support data with inexact matching. The problem of building query engines to execute vague queries has been already studied, but an important problem still remains open, namely that of defining the threshold to be used when a similarity scan is performed over a database column. From the bibliography it is known that the threshold depends on the similarity metrics and also on the set of values being queried. Thus, it is unrealistic to expect that the user supplies a threshold at query time. In this paper we propose a process for estimation of recall/precision values for several thresholds for a database column. The idea is that this process is started by a database administrator in a pre-processing phase using samples extracted from database. The meta-data collected by this process may be used in query processing in the optimization phase. The paper describes this process as well as experiments that were performed in order to evaluate it.

Automatic threshold estimation for data matching applications

“…Further, the experiments demonstrate how the proposed automatic approach leads to results that are close to those obtained by the approach that we developed previously [33], which requires human intervention.…”

“…In previous work [33], we proposed a procedure for reducing human intervention. In that procedure, instead of generating the clusters manually (Step 2 of the procedure above), the human expert just informs how many distinct real world objects are represented by the instances in the samples taken from the dataset.…”

Measuring quality of similarity functions in approximate data matching

“…However, it is worth pointing out that this sampling process could be automated through the use of clustering algorithms, as done in our previous work (Stasiu, Heuser, & da Silva, 2005). In this case, all the elements of a given cluster are considered as representing the same real world object.…”

Automatic and online setting of similarity thresholds in content-based visual information retrieval problems

“…Based on collection samples, a semi-automatic approach for the estimation of recall and precision values for various similarity thresholds minimizes efforts involved by static similarity threshold definitions [28,29]. It requires expert input only where the number of distinct objects contained in each sample is concerned and uses two techniques to reduce human interaction, namely (i) sample use and (ii) similarity cluster process.…”

“…A new approach [5] combines two strategies to eliminate human intervention [28,29], during the recall and precision values estimation process. They are (i) use of agglomerative hierarchical clustering algorithms and (ii) use of the silhouette coefficient for cluster evaluation.…”