Humans and machines working in harmony
Consumer research at scale and in real time is a very attractive proposition for many marketers. Having so much data at hand is exciting; the main issue is you need to know what to do with it to get to be able to derive insights from it. In this post, we explore the role that machines have played, still play and might play in the future of social media intelligence.
Machines and insight
There are many arguments out there about what the word ‘insight’ means. We take it as meaning a fresh and not yet obvious understanding of customer beliefs, values, habits, desires, motives, emotions or needs that everyone agrees on and can become the basis for competitive advantage. Insight is subjective because it comes from the interpretation of data based on the context and culture, which a machine is not (yet) able to do because contextual and cultural understanding is innate, specific and infinitely variable. It can’t therefore be easily programmed.
Knowing what the data means and what the implications are for an organisation is what matters. A lot of data is displayed, but insights are few, and considerations for action are even fewer. A lot of so called social media intelligence out there is about the overall numbers and not what they mean. The people behind the numbers are often forgotten. Consumer opinions are reduced to data points.
The term data mining perfectly describes the role of machines in helping humans develop social media intelligence. They do some of the heavy lifting. They help clear the rock; the ones that do not have any value. They help get to the strata of rock that might yield the ore.
In other words, they help de-clutter, organise, structure and categorise social media data.
The most advanced techniques can help uncover specific themes or trends in the data but you still need humans to help them learn how the content is structured at the outset so that they can read it. More importantly, you also need humans to dig underneath these themes to interpret what they mean for the people commissioning the research. You also need human interpretation to form considerations around how to take action in the real world from what you have actually learnt. To go back to the mining analogy, you need to transform and refine the metal ore to get to the precious metal. That is why machines and humans are both needed to derive actionable social media intelligence.
There are two additional ways machines and humans can work together to make sense of social media content: machine learning which includes natural language processing, mentioned earlier, and human-lead, keyword-based categorisation.
The first sees machines analysing patterns in the content based on an understanding of the structure of conversations. So what humans do is train the machine to understand the structure of sentences to help machines categorise what the content includes.
The second consists in using qualitative analysis done by humans to help devise frameworks that machines can then apply at scale.
Humans and machines working together
We are making the case for the use of humans and machines together because the automation of cumbersome, repetitive tasks releases space for human creativity and interpretation which is needed to derive insights.
Another key example of why humans are crucial in the analysis of social media data is that while tools can determine numerical influence, they cannot profile or segment audiences effectively or even interpret abstract concepts such as advocacy or affinity.
So what does the future hold for the partnership between humans and machines? How can machines help humans in the sphere of social media intelligence in the future?
The first is leveraging the increasing importance of images as a source of insight.
With the rise of Instagram, Vine and continued spread of YouTube, image-based content has become increasingly key to analysing consumer behavior, especially for FMCG brands, where a reported 60% of mentions online are image-based. In Latin American markets, and Brazil in particular Instagram has a huge role in documenting people’s lives but text based searches only capture 20% of all image content.
Machines have an increasing role to play in helping researchers find and analyse these images. Currently this technology is limited, not surprisingly to simple concepts such as logos. Complex logos such as Starbucks are much easier to find than simpler logos like the Nike swoosh.
The future which arguably is happening now already is to have machines analyse images for other constructs than logos such as moods, facial expressions or the clothes people wear. Some of this technology is a long way off but looks promising. I would argue that even with the ability to analyse images in depth, machines are a long way off being able to interpret the context and meaning for the brand. You also have to put in perspective the use you are going to be making of the images you gather using machines. What will you gain from finding pictures of people on Instagram or Facebook with only the context of the image itself to interpret the reason why this image was posted?
The second is in doing some of the heavy data lifting so that researchers can connect intelligence.
Increasingly, organisations also want an integrated view of the digital landscape. It is no longer just about social media itself but also about linking search insights, website experience and the owned, paid and earned equation.
The age of Intelligence amplification
IA as opposed to AI makes the most effective use of information technology and augments it with human intelligence.
Automated analysis is highly useful for providing a landscape view of the data but a layer of human vetting and digging will always be required.
In a recent article published in the Guardian, entitled Machine Learning: why we should not be slaves to the algorithm, John Naughton argues that while machine learning is wonderful it is potentially biased and quotes a recent study by AI researchers at Princeton and the University of Bath revealing that even everyday speech has embedded biases of which most of us are unaware, discrediting machine learning.