Relation Extraction & Parsing

... the process of converting raw text into a set of well structured subject-action-object relationships between entities, phrases and concepts.

TextRazor's relation extraction module leverages our state-of-the-art Dependency Parser and a set of sophisticated linguistic rules to parse relationships in any kind of text. Our approach offers a significant increase in accuracy and recall over alternative solutions, providing you the flexibility to mine an unlimited number of relationship types.

TextRazor's relation extraction system has been used to extract targets of opinions, find management appointments in news stories, extract clinical trial results from medical documents, and parse legal documents.

Dependency Parsing

Typically deep syntactic parsing of language is prohibitively slow and brittle across domains. TextRazor's parser was developed from scratch to tackle these issues head on. We're fast enough to avoid adding significant latency to your requests, and flexible enough to handle different styles of language with ease.

We achieve this with a combination of several systems built on the latest academic research into syntactic dependency parsing, trained on our custom corpus of newswire and conversational English text. The parser includes a ensemble of various globally trained shift reduce, hierarchical, and spanning tree parsers, the output of which are combined and re-scored into the most probable parse.

We've found that this approach yields the greatest accuracy across domains. Since the whole system is heavily parallelized and uses various techniques to speed up processing we can fully parse an average sentence in a few milliseconds, an order of magnitude faster than most alternatives.

Language Support

Relation extraction and parsing is currently only available in English.

API Calls

To retrieve relations from your text simply add the "relations" extractor to your request. For dependency relations only you can use the "dependency-trees" extractor instead. Dependency relations do not have their own response type, instead returned as a property of the word objects.

Read more in our Python Client or REST Documentation.

You can try out the Relation Extraction system with your own documents through our Online Demo.