Voxpopme speeds video-based market research with Watson Natural Language Understanding

Market researchers who work with video feedback, rather than text-based survey responses, spend hours analyzing video footage. Imagine if they could not only understand what respondents in the videos say but also what they think and feel about your products and services, instantly, without watching even a single second of video. That’s what we do at Voxpopme, with the help of IBM Watson.

As one of the world’s leading video insight platforms, according to the 2018 Grit Report (1), Voxpopme helps businesses understand ways to better engage customers and gain keen understanding into their thoughts, feelings and overall loyalty. We do this by ingesting consumer-recorded video and long-format content from focus groups, and then analyzing that data to deliver marketing insights in a wide range of formats. These include graphs, browsable themes of common words or phrases discussed by survey responders, and customizable show reels, which are short videos comprised of whatever story or trend the user would like to highlight.

Our goal in automating the manual work of reviewing video content, is to reduce and even remove barriers separating brands and customers. We chose to add IBM Watson Natural Language Understanding (NLU) to our platform because NLU allows us to identify and aggregate the sentiment of survey respondents, color-code it and then overlay it directly onto transcripts, giving firms better insights to the hearts and minds of their customers instantly and reliably.

With this information our users are able to not only observe what their consumers are talking about, but how those consumers felt about each sentence they uttered – all at a glance. Knowing that consumers are speaking negatively about your recent television commercial is far more useful than just knowing that consumers are talking about it.

AI as the next step in our continuous evolution

Having experimented with sentiment analysis before, we knew that crowdsourcing sentiment in video surveys gives our clients valuable information about consumer consensus. While this works great for short responses, it doesn’t scale well for longer videos – like focus group responses which could run two to three hours long.

These long-format videos contain hundreds of opinions from several different individuals, making a per-response sentiment score useless. We needed quick turnaround times and scores down to the sentence level. For this, only a machine solution would suffice.

NLU helps drive valuable insights, instantly

We developed a unique piece of technology called Theme Explorer, which identifies and extracts the most common talking points among unrelated respondents in a study. It is the part of our platform that allows our users to see, at a glance, what topics survey respondents are talking about (and which ones they’re not). In order to find a solution that would make sure our users didn’t even need to look at transcripts to get valuable data, we trialed sentiment analysis tools not only from IBM Watson but also from Google Cloud and Amazon Web Services. All of the services that we trialed provided fairly accurate sentiment scores for our test data.

Only IBM Watson allowed us to specify the exact terms we felt were important, enabling us to get accurate sentiment scores for every word or phrase that our Theme Explorer technology had deemed to be relevant.

The ability to overlay sentiment analysis on top of each theme makes the information easy to parse in seconds:

Color-coding themes based on their aggregate sentiment scores allows our users to see, at a glance, that 75% of people have spoken positively about the quality of their product, but 87% of those same people have spoken negatively about the product being expensive. To a researcher, that kind of fast insight is invaluable.

With IBM Watson powering our sentiment analysis, we can fundamentally change the way researchers gather data from video surveys.

One year in

Since launching, over 70% of research projects being run through our platform have IBM Watson-powered sentiment analysis as standard.

In addition, we have been able to leverage our sentiment data to build new features that continue to thrill our users, such as extending our show reel generator to automatically create unbiased show reels based on common themes, using the sentiment data to ensure an even split of positive and negative opinions.

We continue to discover new ways to use this data all the time, and our use of IBM Watson is only growing.

1. https://issuu.com/researchshare/docs/grit_q1-q2_2018_final_report

Learn about the AI solution that enabled Voxpopme to achieve success