Extracting RDF Triples from Raw Text

This manuscript manifests the results of our work on extracting RDF triples from raw text data. We took a corpus of news articles and applied several methods for extracting "subject - verb - object" relationships from texts. The first method that we used is syntactic parsing and constructing triples from nsubj and obj syntactic relations. The second approach incorporates extracting semantic roles with the help of frame parser and constructing triples from agent, patient and predicate relationships. After applying the aforementioned methods, a manual evaluation was done over a small number of hand labeled samples. Our system achieved a Precision score of 0.34, Recall of 0.46 and a combined score of 0.39 for F-measure as the final result.