Building a recommendation engine with SAS
If you visit a discussion topic while signed in on the SAS Support Communities, then you've experienced a little bit of AI at work from SAS.
We've recently unveiled a content recommendation engine, built using SAS analytics and model management technologies that are part of SAS Viya. You can learn more about how we built this engine by watching this interview with @Jared, a Principal Data Scientist at SAS. (Click the maximize view icon in the video thumbnail to see the full detail.)
How "personal" are these recommendations? Think of these as suggestions for "topics to explore." The analytical model scores your viewing history against thousands of topics and hundreds of thousands of other users. The result: topics that people "like you" have viewed, but that you have not seen yet (or in a while).
This live engine is a perfect example of the Analytics Life Cycle in action, including:
- Data collection and preparation
- Model development using machine learning algorithms (FACTMAC, or factorization machines, in this case)
- Packaging the model in an analytics store (ASTORE) and wrapping in an API
- Deploying the model and API in a lightweight Docker container, and orchestrating in Kubernetes
- Provisioning access to the API using Apigee, an API gateway
- Monitoring performance of the engine using Apigee, adding collaboration/alerting with Microsoft Teams
- Collecting performance data about click-throughs, and feeding that back into a SAS Model Manager project to forecast and improve the algorithm.
More details and screenshots are available in this article at blogs.sas.com.
It's a great case study for how raw data combined with SAS analytics can result in a benefit, such as an improved visitor experience. Well, at least we hope it's an improved experience. Will the recommended topics be helpful to you, or at least interesting? We're interested to hear what you think.
Not sure that this gives me a lot of confidence in SAS AI ... 3 of the 4 recommendations SAS AI gave me had titles in Korean (that I can't read). I clicked one of the links and the page was also in Korean (with no 'translate' button or anything). So, why is SAS AI recommending pages that are in Korean-only to me? Here's a screen-capture:
