IBM Watson Analytics (discontinued)

We use IBM Watson analytics to replace our data science roles – primarily within social media. I started using it as a recommendation from a client, to get better insight (they're shifting all their software over to IBM. I'm sorry I cannot elaborate further on that). This program is basically a smart data/visual analysis that helps its clients discover patterns within data they upload, for their research. In this case, with our clients, we use it to determine their traffic through their Twitter and Facebook marketing/advertising. It helps us discover what ads are the most responsive and what content resonates best with customers.

For the past year I've run a personal website on auto-pilot that runs a few thousand API calls per month against IBM Watson Alchemy APIs. Hourly, I fetch articles from HackerNews, and feed the subject line to Watson API so it will auto-categorize them for me. Users of my website are then able to browse articles that are more relevant to their interests.

I personally build tech demos to showcase emerging tech and try to encourage others to do so. IBM Bluemix was great for me to bring in APIs like The Weather Channel data and text to speech to create a virtual assistant. I wrote about it online to teach others how to do so too.

Visual recognition demonstration for our customers. We used it for this single demonstration, which took only about a week to 10 days, and the results were presented to several levels or management and we well as 3 levels of customer management with VERY positive response. We trained the the Watson visual recognition tool with 1250 'positive' images of five different objects (250 training images for each object), along with 125 negative images (25 negative images for each object). Then we tested the visual recognition software with pictures of the same type of objects (but using none of the training images) to see how well, and how fast it performed. The positive hit rate was very good (typically 70% or better) with images of the same object, and the rejection rate was typically very good as well (again, better than 70% when the trained object was not in the image). It also did very well when we combined 2-3 objects in the same image. The really surprising thing was that the images could be screen shots (not terribly surprising perhaps), but we could also take a picture of the screen shot with a cell phone and use that--with almost identical results. (When these images were blown up they were quite grainy, so that's why we were surprised.)

However, the problems we encountered with billing, technical support (or total lack thereof), quickly demonstrated that this service was not something we could use for serious work or production.

So I used IBM Watson Analytics in a class that I was offered at my school. It was being used by our department only. It was used to give students an understanding about new technologies in the cognitive computing world. I was also lucky enough to be able to do a project on the same platform.

I am testing out a variety of analytics platforms to get a better understanding of which product is best to use. I typically input ad campaign data and focus on a variety of metrics to see causation and correlation between the marketing efforts and the client company's performance (website, revenue, etc.). This is a great way to verify whether or not advertising is working towards improving KPIs and figure out where to focus the spend on.