AlchemyAPI Updates API, Brings Deep Learning to the Masses

Amy Castor
Jul. 25 2013, 09:00AM EDT

AlchemyAPI wants to make a type of artificial intelligence known as deep learning accessible not only to giant corporations, but to the public. Over the last year, the Denver company has been weaving the technology into its text mining API, recently announcing updates in the areas of relation extraction and named entity extraction.

Launched in 2008, AlchemyAPI crunches through hundreds of billions of words of text on the Internet to delve meaning out of what people are saying, doing, thinking and writing about. The company offers its service, which developers use for media monitoring, ad targeting, surveillance and more, in the form of a RESTful API or a behind-the-firewall appliance.

With its API receiving three billion calls a month, AlchemyAPI is a member of the ProgrammableWeb API Billionaire’s Club. In February, the natural language processing company received $2 million in seed funding to further integrate deep learning into its product. The company says it wants to make this type of artificial intelligence available to anyone who needs it.

“We are trying to democratize artificial intelligent technology so it's not just the Googles, IBMs and Apples that have access to this stuff, but every startup, every university, every medium-sized business in the world,” said Elliot Turner, founder and CEO of AlchemyAPI.

If you are still unclear as to what deep learning is, it is a set of machine learning algorithms that can perform human activities such as seeing, hearing, reading and listening.

Recent advances in the field are taking many industries by storm. Apple uses the technology for its Siri voice recognition system, and Google uses it for image recognition. Chinese search giant Baidu recently built an entire deep learning research facility in Silicon Valley, which it hopes one day will be on par with the Xerox Palo Alto Research Centre. Deep learning was also named one of the top 10 break through technologies for 2013 by MIT Technology Review.

The challenge for deep learning algorithms is to crunch through huge quantities of data and sort elements of that data into categories without knowing in advance what those categories might be. Deep learning algorithms are capable of unsupervised learning. They group things together based on similarities, and from there learn more categories, attempting to figure out lower level categories such as letters first before moving on to higher level ones such as words.

AlchemyAPI now incorporates deep learning algorithms in its relation extraction engine. This feature zeroes in on entities and actions occurring between them, such as a lawsuit between two companies or one company acquiring another.

A new named entity extraction engine, also based on deep learning, now identities people, companies, cities, geographic features and other entities with more accuracy. Within this tool, an improved disambiguation feature examines surrounding text to determine whether, for example, Paris refers to Paris, France or Paris, Texas or whether the Tim Cook being talked about is the CEO of Apple or the ice hockey player. A quotation extractions update extracts quotes from a document and assign the quote to the person who said it.

“While Microsoft and Google are using deep learning, AlchemyAPI is the first to make it accessible to the entire world,” said Turner.

Amy Castor Follow me on Google+

Comments