Relation Extraction Systems

A relationship Extraction System detects whether there is a semantic relationship, usually of a certain type, between two or more given entities within the context in which they appear. For instance, if we want to detect if there's a relation <NaturalDisaster,Location,Date> in the sentence "Hurricane Patrick slammed the coasts of the Cook Island, located east of Australia, on February 8th instead of February 9th as it was predicted" among the entities NaturalDisaster(Hurricane Patrick), Location (the Cook Islands, Australia), Date (February 8th, February 9th), a relation extraction system should return <Hurricane Patrick, the Cook Islands, February 8th>.

The Relationship Extraction task has been studied in different research areas such as Biology, Machine Learning and Natural Language Processing; however, there has been no interest in unifying the implementation of relationship extractor systems. This lack of effort hinders the distribution, implementation and evaluation of these systems, even though they count with an intrinsic similarity. We try to overcome to these issues by developing and providing REEL, an adaptable and easy to use platform for projects of this kind.

What is REEL?

REEL is a Relation Extraction framework for java that allows its users to implement different relationship extraction systems in a few and easy steps. REEL is, above all, an effort to bring different relation extraction systems and tools alike into one controlled and unified environment.

With REEL you can:

·         Extract relations of many different types.

·         Combine documents from different collections.

·         Manage different document file formats.

·         Tokenize document with either state-of-the-art or personalized tokenizers.

·         Integrate your entity taggers and co-reference resolutors.

·         Enrich text with either state-of-the-art or personalized feature extractors.

·         Create your own statistical structures (e.g., kernels) or start with the ones we provide.

·         Incorporate your favorite engine libraries (e.g., jlibsvm) to process your structures.

·         Train your models in one step.

·         Evaluate your models using state-of-the-art metrics and others specific of your task.