Case study: Ocado uses appurtenance training to assistance patron services understanding with email overload

A healthy fit

Ocado has been regulating a Google Cloud Platform (GCP) for big information analytics projects given early 2014, and primarily deliberate regulating a internet giant’s Cloud Natural Language focus programming interface (API) for this project.

The API is used by developers to lift out view research and remove sum about specific things from datasets, formed on believe it has accrued from trawling a internet.

“It is a good API, though it is lerned around internet information rather than information specific to a business problems,” says Nelson. “So, while it took us a satisfactory way, it couldn’t solve everything.”

This stirred a association to start regulating a open source appurtenance training toolkit TensorFlow, paving a approach for it to build bespoke appurtenance training models of a own.

“The group went away, looked during opposite postulate that would concede them to solve a healthy denunciation problem and radically replicate, though during a vast programmed scale, a triage and tagging of patron emails,” he says.

The Ocado appurtenance training indication was “trained” by feeding it with 3 million genuine patron emails that were hosted in Google’s open cloud.

“It is a unequivocally changed dataset for us, and allows us to make a indication really accurate to a tinge and context of a customer,” says Nelson.

“We didn’t know if it was going to take one practical appurtenance to plough by 3 million emails or a million machines, and regulating Google’s cloud, it took divided any worries we had about either or not we had adequate hardware capability to do this.”

The indication is now deployed within a hit system, where it is used to consider either a tinge of incoming emails is positive, disastrous or neutral, and tags them so a patron use group knows how high a priority they are.

“It is trained, and now it is in a training state where we’re creation certain it works during full scale and measuring a success rate,” he says.

Article source: http://www.computerweekly.com/news/450401505/Case-study-How-Ocado-is-using-machine-learning-to-help-customer-services-avoid-email-overload

Related posts