How to uncover your consumer’s aspirations in less than a week
FMCG companies are using digital and social media tools to observe and to “listen” to what consumers say about their products. But what if you had to capture insights about their goals and key drivers?
Such insights relate to more personal and less openly claimed data. This results in market researchers tending to put less faith in social media as a source of valuable insights. It is also a popular belief within the Market Research community that social media data is too biased to be considered as a reliable source for answering complex questions (The problem with social media monitoring tools).
While these objections are very legitimate, recent advances in Artificial Intelligence (AI: the future is now) allow for the mitigation of biases like these. It helps to filter data, and more importantly, to identify the correct target audience of consumers.
Recently, at Beautifeye, we developed a methodology that helps to answer complex questions such as: “How and why do young Argentinian women express their of sexuality online?”
Our research has revealed the answer to this hypothetical question in the methodology detailed below:
1. Build the right dataset
The key to successfully extracting insights from unstructured data is to aggregate the correct type of statistics from the desired sample group of people. Next, we determined a platform on which to gather said data. In this case, we chose Instagram. To build the right dataset, we implemented the following three steps*:
- Sample robustly: we created a panel of 13,000 Argentinian women and their biographical data. The resulting corpus included more than ½ million images and related text.
- Model your data: we generated an ontology of keywords from posts captions in our sample group while simultaneously filtering irrelevant keywords.
- Establish a baseline: we created a second “baseline panel” of roughly the same size of the group in step one. This “baseline panel” aggregates women in the UK.
*More details about the panel creation are included in the full report.
2. Filter Noise with Image Recognition
In order to remove the overwhelming amount of noise in the data, hashtag filtering is not enough.
To focus on relevant content, we used a filtering step based on visual analysis. In particular, we were able to automatically detect pictures of women, and discard other irrelevant data.
It is important to note that by using image recognition combined with text analysis we got rid of 71.4% of irrelevant posts from our corpus.**
**The accuracy of the visual analysis was 82%, which we discuss in more detail in the full report.
3. Test your hypothesis
Equipped with a solid baseline, a noise free corpus, and rich visual data, we were finally ready to answer our fundamental question: “How and why do young Argentinian women express their of sexuality online?”
Our key finding was pretty surprising:
“Argentinian women are twice as likely to express their sexuality on Instagram than British women.”
A second finding that emerged in the qualitative analysis of the data on visual dashboards is the following: Argentinian women are openly expressing their sexuality in a different way than what is portrayed in the mainstream media. It can be expressed through tattoos, piercings, subcultures (e.g. goth), bold makeup or very colorful hairstyles. By making these style choices, women do not deny their femininity. Instead, they express it in a non-traditional way.
Our objective was to show you that with the right approach and the right technology, social media can address complex, cultural and personal traits of consumers. We outlined here one of the many methodologies that can be implemented using Artificial Intelligence combined with visual social media. While traditional research approaches are still the only way to dive deeply into such insights, novel approaches such as this one can be used to guide the search. In doing so, a considerable amount of time and effort can be saved.
You can download the full report containing the completely detailed methodology here.