Combining Linguistic and Statistical Analysis to Extract Relations from Web Documents

abstract 1: The World Wide Web provides a nearly endless source of knowledge, which is mostly given in natural language. A first step towards exploiting this data automatically could be to extract pairs of a given semantic relation from text documents - for example all pairs of a person and her birthdate. One strategy for this task is to find text patterns that express the semantic relation, to generalize these patterns, and to apply them to a corpus to find new pairs. In this paper, we show that this approach profits significantly when deep linguistic structures are used instead of surface text patterns. We demonstrate how linguistic structures can be represented for machine learning, and we provide a theoretical analysis of the pattern matching approach. We show the benefits of our approach by extensive experiments with our prototype system LEILA. abstract 2: Search engines, question answering systems and classification systems alike can greatly profit from formalized world knowledge. Unfortunately, manually compiled collections of world knowledge (such as WordNet or the Suggested Upper Merged Ontology SUMO) often suffer from low coverage, high assembling costs and fast aging. In contrast, the World Wide Web provides an endless source of knowledge, assembled by millions of people, updated constantly and available for free. In this paper, we propose a novel method for learning arbitrary binary relations from natural language Web documents, without human interaction. Our system, LEILA, combines linguistic analysis and machine learning techniques to find robust patterns in the text and to generalize them. For initialization, we only require a set of examples of the target relation and a set of counterexamples (e.g. from WordNet). The architecture consists of 3 stages: Finding patterns in the corpus based on the given examples, assessing the patterns based on probabilistic confidence, and applying the generalized patterns to propose pairs for the target relation. We prove the benefits and practical viability of our approach by extensive experiments, showing that LEILA achieves consistent improvements over existing comparable techniques (e.g. Snowball, TextToOnto)