Collective Information Extraction with Relational Markov Networks

Most information extraction (IE) systems treat separate potential extractions as independent. However, in many cases, considering in uences between dierent potential extractions could im-prove overall accuracy. Statistical methods based on undirected graphical models, such as conditional random elds (CRFs), have been shown to be an eective approach to learning accurate IE systems. We present a new IE method that employs Relational Markov Net-works (a generalization of CRFs), which can represent arbitrary dependencies between ex-tractions. This allows for \collective informa-tion extraction " that exploits the mutual in-uence between possible extractions. Experi-ments on learning to extract protein names from biomedical text demonstrate the advantages of this approach.