On Precision and Recall of Multi-Attribute Data Extraction from Semistructured Sources

Guizhen Yang    Saikat Mukherjee    I. V. Ramakrishnan

Abstract

Machine learning techniques for data extraction from semistructured sources exhibit different precision and recall characteristics. However to date the formal relationship between learning algorithms and their impact on these two metrics remains unexplored. This paper proposes a formalization of precision and recall of extraction and investigates the complexity-theoretic aspects of learning algorithms for multi-attribute data extraction based on this formalism. We show that there is a tradeoff between precision/recall of extraction and computational efficiency and present experimental results to demonstrate the practical utility of these concepts in designing scalable data extraction algorithms for improving recall without compromising on precision.