One new feature is dubbed Relation Extraction, which project engineer Shaun Roach tells the Semantic Web Blog “detects sentences describing actions, events, and facts, and then codes them into a machine-readable format. It is a key feature for developers who want to go a step beyond tagging, to understand specifically how all the people, places, and things mentioned in the document are interacting.”
So, it processes natural language, and converts documents and web pages into actionable, semantically enriched “Subject-Action-Object” data, as the company blog describes it.
Roach says the capability is of interest for various customer use cases, including enhancing business intelligence where massive amounts of content must be analyzed, and changes in the structure of an organization or in some other object pinpointed.
As an example, he explains how a recent press release announcing “Edd Helms joins International Business Brokers,” can be run through AlchemyAPI, with output that “sees that Edd Helms has joined International Business Brokers. Further, it sees that Edd Helms is a person, and International Business Brokers is an Organization. If the user of AlchemyAPI wanted to quickly analyze a large volume of documents, and sift out just the relations where a Person ‘joins’ an organization, this functionality makes that easy.” For many entities the service has extra subtypes, he adds.
Or a publisher could use the API to derive from the sentence “Michelle Bachmann visited Iowa on Tuesday,” the fact that Bachmann is a recognized politician and Iowa is a state, and the two are joined by the visit there. It could take advantage of such entity and relationship understandings to pull together all the articles about politicians’ swings through midwestern states during some period, and deliver that information to readers in an interesting fashion, perhaps as a visualization.
“There are many other examples,” he says. “Companies looking at Twitter would want to tag anyone using some form of the verb ‘buy’ in association with their product,” for instance.
The service also has enhanced its sentiment analytics capabilities to include directional sentiment analytics – that is, that one entity is emitting negative or positive sentiment to another entity mentioned. “The relations extraction makes the directional sentiment much more explicit,” Roach says. “The software determines if an entity is ‘sending’ or ‘receiving’ sentiment.”
Take the sentence, Mitt Romney criticizes Obama economic policies: “We’ll pull out the relationship of Mitt Romney criticizing Obama economic policies. And there is a new sentiment tag “sentiment from subject”, which is negative, and which indicates that the subject (Romney) is sending negative sentiment towards the object (Obama Economic Policies),” he says.
“So, whereas before you would potentially get a negative sentiment associated with Mitt Romney, that sentence doesn’t really express negative sentiment towards Mitt Romney. It expresses the fact that Mitt Romney has taken a negative action against Obama economic policies.”
In May the natural language processing platform added some other features, including new co-reference resolution capabilities to resolve referent mentions within a document back to their source entity, and improved entity disambiguation functionality for increased accuracy and faster performance.
- Big Data Skills Worth Big Bucks
- Automatic Hashtags & Machine Learning: The New Google+
- Cambridge Semantics Wins SIIA Software CODiE Award
- NLP Company Versus IO Raises $2.8M