The youngest children in the AI revolution are language models. They go by fancy muppet names such as ELMo (in 2018, by the Allen Institute for AI), BERT (in 2018, by Google) or RoBERTa (in 2019, by Facebook AI). These language models are deep neural networks trained in simple linguistic tasks on vast amount of data (usually nothing less than all published novels and all of Wikipedia). What they provide are representations for sentence meanings that can be used for pretty much any semantic task, ranging from semantic search to text classification or information retrieval.

We would like to introduce a market research application for those language models, which allows us to identify hot topics and matters of interests among any pre-defined target. More precisely, we will present an online tool that allows the user to query for any topic she is curious about (say: “climate change” or “shameful Harry”) and get the reach of that topic among various audiences defined in terms of socio-demographics as well as political leanings (are conservative voters really indifferent to climate change? do younger folks really care about Harry?).

Our tool works on the basis of a content-driven analysis of the internet usage of respondi’s online representative panel (people having installed a tracking device on their laptop and mobile). Contents and queries are matched using fine-tuned neural networks on top of Google’s earlier mentioned muppet magic for natural language processing, BERT.

We think this approach shows one way for the future of opinion analysis, by using AI and language model in order to directly research people’s interests and concerns as inferred from their web navigation. This approach is (almost) real-time: queries dig into the most recent web behavior of our panel members (D-1 and earlier). It is (completely) open-ended: topic and issues are not a closed list, one can ask for any topic. And it relies on a simple but crucial crossing of data sources: traditional survey data such as age, gender or political preferences allows for the definition of targets groups whose specific interests one cares about.

Iain Mackay and the X-MR team draw on years of experience in working with survey data across a plethora of platforms and languages.  They will explore the topic of transferring survey data into the world of relational databases, highlighting that many big problems have long been familiar:

• Representation of multiple classification variables and grids
• Using value codes as well as labels
• The shape of data
• The rate of change of metadata
• Communication issues – different terminology for the same concepts in the different domains.

Solution

They will begin by providing a phrase book, to help those travelling between the survey and corporate data communities.

They go on to demonstrate deploying a substantial consumer survey, from survey format into an enterprise query tool, via a relational database.

They explain the problems inherent in the process, and their solutions. Although the demonstration is practical and specific, they will expose the methodology in a way that is actionable and vendor neutral.

No matter how robust your methodology, what really matters in the end is that clients are able to relate to your findings on a personal level and connect with what they mean for future strategy.

In this presentation, Dr Matilda Andersson, Managing Director at Crowd DNA London, will share how to authentically connect, empathise and represent people through immersive methodologies. Standard demographics and cookie cutter segmentations no longer work; categories are dissolving as audiences refuse to be boxed in; and people seek to be in control of their own image, irrespective of gender, sexuality, race, size or age. This means we need new methodologies. But also be able to communicate research findings in a compelling, yet commercial way.

Dr Andersson will explain how we achieve this via techniques such as IRL immersions, social media ethnos, multi-generational in-depth interviews, and putting control back in the hands of participants (to turn them into partners, not respondents). She’ll also explore the need to produce narratives that move beyond traditional formats – think curated Insta feeds, Snap stories, child’s-view GoPro footage, street photography – and how, in order to hold onto the grit of the core story, we need creative planning, careful execution and organic formats to help brands truly empathise with their audiences beyond survey results.

The data processing and analysis software eco-system grows richer by the day, but too often where we want a bridge, we find instead a “one-way” railway track.

To address this issue, the Red Centre Software team have developed a stand-alone cross-platform survey tabulation and analysis tool (implemented as a zero-GUI DLL), for bridging between any data set and downstream interactive visualisation tools.

As a proof of concept, they will use social media data collected the public Twitter API on a variety of current affairs.  The reformatted case data is tagged with sentiment scores and text analysis, then analysed and visualised using familiar concepts: bases, row/column percentages, filters, weighted/unweighted, multi-response, grids, coded increments, standard crosstab statistics, moving averages, etc.

Generalising the notion of “a respondent” is not just limited to tweets, it can include any set of periodic readings, or data sets, which are the traditional preserve of BI and SQL.

This paper describes the required DLL functionality to fulfil the role of a practical and effective bridge, and details the data transforms necessary to effect cross tabulation.

Organizations rely heavily on consumer behaviour and data for decision making. However, their ability to access consumer data has become increasingly difficult, restricted, and often inaccessible with the rise of mobile phones, apps, privacy protocols, and consumers’ reluctance or lack of trust to share information.

This session will focus on how non-survey data (behavioural), in new forms, can provide new data opportunities outside of surveys, and also further contextualising the information captured in surveys, all within the evolving privacy-first environment.

This session also explores behavioural data compiled from more than 5,000 MSR app users. Specifically, it explores what people are doing (e.g. app usage), consuming and watching (e.g. streaming services and social platforms) and buying (detailed Amazon purchase history), and explores the opportunity the new technology provides to bridge surveys to a wider world of behavioural data.

Abs Chamali and the ConceptSauce team have built a VR focused platform, using opensource technology, that is adapted for Market Research.  They will be sharing many of the challenges, considerations, knowledge gained, and features developed through that journey.

As they say, “A key challenge for us was to find the right set of technologies that would work together efficiently.

Our clients’ needs for features, from VR to special applications, meant that we had to collate a set of technologies that would allow us to adapt, build solutions quickly, and with stability.

We have experimented with many technologies and overcome a mountain of challenges. We will run through the issues faced on the backend, the frontend, the architecture and other services.

We’ll share how we use APIs to connect data with other MR platforms.  We’ll also demonstrate our latest platform using nodes as data sources, and showcase how this enables the user to build standard shopper shelves, advanced conjoint-based planogram arrangements, and VR deployments.”

Matt Bishop and Dave Morris will draw upon recent case studies to share some useful hints and tips about using WhatsApp and other messaging apps for automated survey research. Attendees will learn what messenger apps such as WhatsApp can bring to insight, speed, scale and cost of doing survey research with different audiences; from business owners in the UK to students in Brazil.

Matt and Sue will also show demos of Jigsaw’s unique WhatsApp approach, and discuss its applications. Implications for the future of digital survey research will be considered, and attendees will also have the chance to scan a QR code to try the tool out for themselves.

Addressing the need to make mobile surveys more natural and inclusive for respondents, Jigsaw Research has become the first agency to offer a fully automated WhatsApp approach. This solution won the MRS ‘Innovation of the Year’ Award 2020.

An update from the TSAPI Steering Group on developments in the creation of a new standard for survey data transfer.  The aim is to create a standard relevant to a world which is increasingly defined by interconnectivity across all platforms and devices.

Overall objective are:

• Help preserve & develop MR’s position at the heart of businesses intelligence.
• Help improve connectivity within the MR industry.
• Help facilitate innovation within the MR industry.
• Help ease the connection of MR insights & value to other sectors.

We’ll be inviting critique and collaboration to help achieve these goals.