Making sense of unstructured text and video data
The explosion of Big Data has brought an increasing focus on how our clients manage and drive value from their data assets, however less focus has been applied to the growing volume of text and video outputs from research. Tim Osborn, an Associate Director at Verve, has been taking a lead on ensuring we make the most of these sources via innovative, and rapid, approaches to analysis.
One message we hear consistently from clients is that they need information “yesterday!” and, while we’re not magicians, we take this seriously when it comes to develop on-going, successful partnerships. As a result, aligning the pace of our business with those of our clients is a top priority for Verve.
Starting to apply this mindset inevitably means moving beyond traditional approaches and entrenched quant-qual mindsets. All too often, the unstructured data we collect (text, videos, images) is too time consuming and numerous to analyse case by case, but not neat enough or as easily codable as a simple percentage measure of preference. So, how can we work nimbly in this middle ground (i.e. the real world) where things don’t fit neatly into methodological buckets? Imagine a scenario where you run a pop up community with 500 people to tell a story about their tea drinking habits, or collect 15,000 open ended responses linked to NPS data on in store experiences. Both scenarios would generate a lot of information and, with tight deadlines, there often isn’t time to sift through everything individually.
At Verve, working with fast paced clients means insights are sometimes needed quickly without the luxury of long lead times and analysis sessions. We have recently looked to text analytics as a way of optimising the often time consuming analysis and reporting phase – a way to sift through large volumes of unstructured data and help researchers quickly home in on the key themes of interest. Whilst the technology is still in its infancy and no match for a researcher’s skill at understanding context and meaning, we can quickly classify sentiment and pull out key themes at an overall level – guiding the researcher to look in more depth at specific groups or run follow up analysis. As natural language processing becomes more advanced, it’s inevitable that machines will become better at understanding context and detecting subtle themes and language commonalities across thousands of responses, whilst simultaneously blending this big picture view with individual customer stories.
Our Verve Videos solution addresses a similar problem researchers face when dealing with large amounts of consumer generated video content. This works for both short vox pops and longer form video, where consumers share key moments in their day or their views around products or brands. Our system automatically transcribes their responses, codes them by sentiment and theme and makes it easy to create a customised show reel ready for clients to share. Building a library of ‘instant opinions’ searchable at a moment’s notice inevitably speeds up the process of gathering and sharing customer voice across a client’s business – and over time this forms a ready to access library of video insight.
Investing time and money on these approaches is already paying off and helping our clients get more from ‘messy’ consumer data. Increasingly, the future is going to be about embracing new technologies and finding new and innovative ways to work with unstructured data – at pace and ensuring we work collaboratively with clients to help them react to changing business needs.