High-Performance Information Extraction with AliBaba
Authors
- Peter Palaga (Humboldt-Universität zu Berlin, Germany)
- Long Nguyen (Humboldt-Universität zu Berlin, Germany)
- Ulf Leser (Humboldt-Universität zu Berlin, Germany)
- Jörg Hakenberg (Arizona State University, USA)
Abstract
A wealth of information is available only in web pages, patents, publications etc. Extracting information from such sources is challenging, both due to the typically complex language processing steps required and to the potentially large number of texts that need to be analyzed. Furthermore, integrating extracted data with other sources of knowledge often is mandatory for subsequent analysis. In this demo, we present the AliBaba system for scalable information extraction from biomedical documents. Unlike many other systems, AliBaba performs both entity extraction and relationship extraction and graphically visualizes the resulting net-work of inter-connected objects. It leverages the PubMed search engine for selection of relevant documents. The technical novelty of AliBaba is twofold: (a) its ability to automatically learn language patterns for relationship extraction without an annotated corpus, and (b) its high performance pattern matching algorithm. We show that a simple yet effective pattern filtering technique improves the runtime of the system drastically without harming its extraction effectiveness. Although AliBaba has been implemented for biomedical texts, its underlying principles should also be applicable in any other domain.
Session
EDBT Demo Session 2: Demo Group 2 (Wednesday, March 25, 14:00—17:30)