This post is authored by Sudhesh Suresh, Senior PM at Microsoft.
Several Azure customers have asked to combine the real time analytics capabilities in Azure Stream Analytics with the power of Azure ML in quickly building and operationalizing any machine learning model as a web service. We are very excited to announce that this feature is now in public preview – you can now apply Azure ML models as a function on top of streaming data to get real time insights.
This release allows you to score individual events of streaming data, leveraging an ML model hosted in Azure. This lets you easily build applications for scenarios such as real time Twitter sentiment analytics, for instance.
Here are two scenarios we are implementing with our customers:
One customer is using this capability to provide real time product recommendations on their website, helping them drive more revenue. Recommendations get served in real time based on website click data, user profile and other contextual information that is being scored against an Azure ML product recommendation model.
Another customer is extracting, in real time, the topics and sentiment associated with conversations happening between their customers and support staff. Support managers use this information to become aware of any critical customer issues in a more timely manner, which helps with customer satisfaction and retention.
To use this feature in the Azure portal, under the Azure Stream Analytics service you’ll see a new option called FUNCTIONS (shown below) which lets you add the Azure ML web service as a function. The ability to get ML scores by aggregating multiple events is not yet supported, but it’s something we are looking into.
Here is a tutorial to get started – the tutorial scenario involves analyzing sentiment on streaming text data.
We hope you have a chance to try this out and look forward to some cool real time applications from the community.
Follow me on Twitter