Our technologies for sentiment measuring opens eyes to our clients

The ability of sentiment evaluation in our surroundings is the key skill for human life. It is a great value also for every marketing oriented company. You will find an ideal indicator for this on Facebook. You just have to choose the right view and be ready to process a little bit of data.

For the human’s social life the value of feedback is uncountable. We are in daily contact with our surroundings. We interact with members of our family, with colleagues at work and with many others during many occasions which fulfill an ordinary day.

The life in a group surely shapes us. We will most probably react to shown traces of unpleasure in our partner’s face and we change our behavior for the future occasions. Unless we would suffer by some personality disorder or even psychopathy. 

When the human met a brand

The brands enter humans social space too. They did it even before the digital era and it’s modern web social networks. And obviously, the brands don’t want to look psychopathic. Nobody sane does want to cause this kind of user experience.  

The possibilities of brands for correction of their behaviour and also the dynamic of a brand’s social interactions (Corporate level) are obviously totally different compared to individuals (Personal level). There could be hundreds marketing campaign outputs and millions are reacting to them simultaneously. To comprehend the feedback in this structure could be problematical.

The marketing departments of big brands are very well aware of that. They build the whole system for measuring and observing mood waves of theirs customers – traditional focus groups, surveys etc. This traditional way is from it’s essence difficult, therefore expensive. Also it is not possible to gain and apply learnings from this approach as quickly as reality demands it. Not even to mention that information gained by a traditional way looses inevitably lots from its granularity – it is complicated to get to know what sentiment had a particular creative at a particular moment.

What is a Sentiment and how to read it from Facebook

According to Wikipedia as well as according to philosophical schools across centuries, the term sentiment stands for general naming of experiencing emotions. A few years ago, Facebook gave to all its users a crystal clear shortcut how to express their emotions. Facebook added a complete palette of human emotions to its simple “Like” button in an option how to rate posts and other content.

Advertisers together with admins of company profiles got from Facebook additional data layers for evaluation of their online activities.

Emotions as a KPI of marketing departments

The practice of Business Intelligence guides us to involve methods of maths and statistics even for matters of fact as transcendent as are aggregated emotions of a fanbase of a brand or target audience. You can find the basic instructions how to catch the sentiment of your audience in the routine described below.

At the first you have to make yourself clear about what kind of emotions you want to invoke with your content. According to your intention you can then rank particular facebook reactions. That shouldn’t be difficult. In general it is possible to say that the emotional vector of reactions should go in the same direction as the content which caused it. We are expecting that a sad story would cause sadness, the joke attempt to cause laughter and so on. 

Sort your feelings 

You won’t make a mistake for most of the marketing objectives, if you sort the reactions in a way shown in the picture below.  

The symbol of a heart is representing Love – the highest good for Christians and other ideologies across the globe. It is hardly possible to imagine a better relationship from the perspective of a customer/brand interaction, than the one which the heart symbol represents.

The opposite side of the spectrum leaves no room for discussions also. An angry customer is from position of the company the worst thing. The sad client is only slightly less bad.

The discussion could be held above the sorting of two – more ambivalent – reactions which left. This creates a grey area which is most endangered by the possibility of irony. Especially laughter – haha – is connected with this phenomenon. But let’s together with Aristotle and his Poetics  justify laughter. Let’s for simplicity assume that it has a higher value than the plain “like”.

 The suggested approach isn’t and doesn’t want to be the only one right and valid interpretation – so feel free to adjust it to better fit your taste and intentions.

Time for statistics

Once you have a scale of the sentiment set up, you can start to observe your content, how it stands on this scale. You get this information simply by the calculation of the average by the reaction values. The more correct and relevant picture of the reality than the plain average offers the weighted average, described by the formula below.

I:) – Reaction Index – weighted average of the reactions value
w – is weight, which is defined by the count of particular reactions
x – is value of reaction, according to set-up scale above

Limitations of reactions and the next steps in the sentiment analysis

What might seem crystal clear at first sight but gets blurred straight after. Every setting, similar to the one shown above, bears the risk of mistake. It assigns the exact numerical value to the particular reaction universally and without taking any context into account. It ignores the possibility of the irony and other difficulties resulting from the complexity of human interactions.

It is at least partly possible to solve this problem with enrichment of the evaluation algorithm by the process which defines the convergence of the published content and caused reaction.

Typically, if we communicate an objectively sad topic, an adequate reaction is sadness as well. In this case it is more valuable for us then laughter, although according the initial settings sadness had a lower value. 

Text classification and machine learning

If we want to exceed plain reaction measurement in sentiment evaluation, we have to cope with the needs of the text (or content) classification. In principle, two paths are leading to this goal. The first path involves a human moderator (better group of moderators). Moderators go through the particular posts and tag the content with the appropriate sentiment label according to their own reasoning. Pros are simplicity and a low danger of misinterpretation, on the other hand cons are a lack of flexibility and high cost for  a unit.

The second path means clustering and classification of text by machine and methods of Natural Language Processing (NLP). If you don’t want to start in this area from scratch, a good point where to start is a page for Facebook open source libraries. For dealing with NLP there is fastText – library for efficient text classification and representation learning, already pre-trained for 157 languages (Czech included).

The ideal solution, as usual, lies in between those two paths. The methods of machine learning provides you with sufficient capacity for processing a higher amount of tasks, while the supervision of a human moderator lowers the risk of misinterpretation within a message. 

Related Posts