IBM Watson Analytics for Social Media

IBM | 2015
UX Team: Aparna Satoor, Mel Rodriguez-Zynda, Kev Bittner

With the rise of social media came the the need to analyize the billions of data points created each day. Each tweet, comment, and review contains valuable data for brands and researchers. Whereas other types of analytics rely on numbers, social media analytics is based heavily on understanding human-written content.

When I joined the Watson Analytics team I knew next to nothing about analytics. As with many projects the learning curve was steep but fascinating. I joined the project to redesign IBM’s Social Media Analytics product. The powerful back-end of the product was masked by the disjointed experience and complicated UI. I think the only people who knew how it all worked were the engineers who built it which made it a challenge for users.

One Product, 3 Disjointed Parts

This part of the product was one of 2 places to go to dig into your data and try and learn from it. It was difficult for users to navigate and near impossible to export charts and graphs for their stakeholders.

I know what you're thinking. "Are those all the same product?" The answer is yes. That was one of the concerns we had as a UX team when kicking off this project.

Through interviews with 18 users conducted by a research team member, we uncovered a few main issues we needed to address.

  • Disjointed experience - The product had 3 very different sections that a user needed to navigate between in order to configure the data and analyze the results
  • The configuration was too complex - It took weeks for a data analyst to refine the model enough to pull the right data

  • It's difficult to see the desired visualizations - filtering through the data is complicated and as a result users can't get the visualizations they need to communicate with their stakeholders and make actionable decisions
We also created a persona of a user who wants to do her own social media analytics, but can't because the Social Media Analytics product is too complex.

This is Andrea's current journey to gain social media insights for the mayor's office. With the product requiring expertise to use it, Andrea must work with an agency to get learn from social media data.

Our MVP Goal

A business user working in Watson Analytics can query and explore social media content to gain relevant and useful insights, without assistance from data scientists or IT.

As we learned more about the way the product is built and how to configure the most effective social media models, we created a list of feature requirements for whatever framework we were going to use for the redesign. This list was created through conversations with our product manager and the engineering team.

  • A user can visualize the hierarchy of topics, terms, and context rules in a manner that illustrates the relationship between them (topic tree)
  • A user can understand how a model is built (ex. topics pull documents, KPIs don't)
  • A user can see how changes to terms, sources, ect. affect the preview
  • A user can view snippets at any time to confirm the model is finding the right content
  • A user can view related terms once the user types in a specific term
  • A user can look at distribution of results across dates, sources, languages authors
These principles helped us to evaluate our sketches and wireframes and lead us in the right direction.

Part of my process is to diagram to help test and adapt my understanding. It proves to be a useful tool to make sure developers, product managers, and UXers are on the same page.

Early Stage Wireframes

A lot of our effort went into thinking about the configuration of the model. It's a laborious process that involves including an excluding specific terms in order to pull the right data. In this idea I surfaced a word cloud to help the user to refine their model. If they want to learn about what is being said about the HTC One phone by walmart shoppers, they may want to include mentions of the HTC One AND Walmart in their model.

Hi-Fi Wireframes

This is the new design for the configuation screen. Here, a user would set what data to pull including languages, keywords (topics), sources, etc.

Here is a preview of some of the results to help the user to refine their data model. The users we spoke to had trouble making sure the app was pulling the correct data - especially since the pulling of data could take hours or days. This preview helps to solve that problem.

Once the data is pulled, a user can show and hide different aspects to find actionable insights.

This is the last work I accomplished for the project before leaving IBM to become a UX designer for athenahealth. As you can see below, the team who continued the project did not drastically change anything when implementing and iterating on the designs.

Implemented Product

Successes

  • Collaborating with engineers and a product manager despite a lack of co-location - we held regular video chat meetings to keep the team on the same page. We even had everyone fly into Austin for an in-person workshop.
  • Consolidating the 3 disparate product sections into 1 in an elegant way.

Challenges

  • Proving the value of user experience design - Our engineering and product peers had little to no experience working with UXers. We tried to keep them as involved as possible in our process to help them understand our role.
  • Changing requirements - At first this product was going to exist as a stand-alone offering, but a few months later it was going to be worked into Watson Analytics. We had to pivot our design ideas a bit, but luckily a lot of our past designs could be incorporated.

Other Projects

Diabetes: What is it?

The Sunday New York Times

One Week

Surgical Scheduling and Inventory