A Learning Classifier-Based Approach to Aligning Data Items and Labels

摘要

Web databases are now pervasive. Query result pages are dynamically generated from these databases in response to user-submitted queries. A query result page contains a number of data records, each of which consists of data items and their labels. In this paper, we focus on the data alignment problem, in which individual data items and labels from different data records on a query page are aligned into separate columns, each representing a group of semantically similar data items or labels from each of these data records. We present a new approach to the data alignment problem, in which learning classifiers are trained using supervised learning to align data items and labels. Previous approaches to this problem have relied on heuristics and manually-crafted rules, which are difficult to be adapted to new page layouts and designs. In contrast we are motivated to develop learning classifiers which can be easily adapted. We have implemented the proposed learning classifier-based approach in a software prototype, rAUgner, and our experimental results have shown that the approach is highly effective.