Synthesys™ provides state of the art entity extraction and categorization from unstructured text by leveraging best of breed algorithms and statistical analysis. In addition to traditional natural language processing which marks up text with part of speech tags such as nouns, verbs, prepositional phrases, and direct objects, Synthesys goes one step farther by deducing which of the words in the sentence are entities — the most interesting parts of the sentence. Additionally, Synthesys takes the analysis to a much deeper level and performs advanced statistical analysis to determine what kinds of entities are in the sentences, how the entities interact with one another, and what the basic nature of the entities might be in terms of ontology.
To illustrate the basic technique, just take the simple sentence, “Jack and Jill went up the hill”. Synthesys first performs natural language processing to determine that things such as “Jack” and “Jill” are nouns in the subject of the sentence, “hill” is an object in the predicate of the sentence, and “Jack” and “Jill” performed some kind of action on the hill, which in this case is that they “went up” the “hill”. However, you wouldn’t want to stop right there — you’d want to take analysis to the next level and deduce semantic level meaning by building out an object graph in which “Jack”, “Jill” and “hill” are categorized entities in the graph with edges describing the proper relationships between them. For this particular illustration, you’d expect something like “Jack” and “Jill” being people and the “hill” being a location that they are somehow associated with. Although this simple example is easy to comprehend at an intuitive level, solving the general case is a much harder problem to solve with high quality results at a reasonable speed.
“Jack and Jill went up the hill” is a piece of text with no semantic level meaning. Compare that to a rich object graph (See Below) in which extractors have identified the most interesting things in the sentence and the relationships amongst them.
Semantic Object Graph example
