Social media and big data have been living a passionate relationship in the past decade, giving birth to a multitude of tools attempting to reveal its secret, for the sake of fruitful marketing operations. Indeed, some studies have been conducted on how to process and make the most of social media data from which hundreds of analytics tools have emerged, promising to provide businesses with fascinating information about their brand online.
How do these tools actually process social data? How can you manage to deduce any meaningful information from this huge unceasing flow of messy data? What are the limits of these methods?
- When we started talking about ourselves
First, it is important to understand how it started and what makes this social data so special. Let’s fly a dozen years back, when customers began to actively share information about themselves, their friends or about new items they bought, which was no longer limited to search and click data (Weigend, 2009). What pushed them to do so? Actually, users became increasingly active in some ways because companies influenced them to contribute by giving them incentives. For instance, because Netflix users prefer relying on other users’ recommendations to choose a movie, it gives them an incentive to contribute and rate movies as well. As a result, consumers now have a voice online and companies realized that this actively generated data is truly valuable by helping them make decisions. In fact, measuring this online voice has become the core business of a special kind of analytics software products, called social listening tools.
- Listen to the world
Now that people are expressing their feelings out loud about your brand and your competitors, it is time to listen to them. In reality, the whole idea here is not only to collect as much data as possible, but also to identify what is really relevant for you, which requires cleaning and selection. Of course, this should be done fast, as close as possible to real-time, which may not sound as easy as ABC. Indeed, according to Internetlivestats.com, 500 million tweets are sent every day on Twitter and around 200 billion per year… and that is only one source! Fortunately, in 2017 there are tools to collect, store, clean and share this data with you.
If we focus on the data coming from Twitter, the only way social listening tools can capture every single tweet in real-time is to be plugged to the “Twitter Firehose”. This is the real-time stream of tweets flowing from the social platform every day, to which only a few trusted partners have access. Synthesio, Brandwatch, Linkfluence or Sysomos are some of the few tools having a granted access to the Firehose.
They have now absorbed and fed all these tweets into their own database. The next step, which requires human action, will allow you to understand your people.
- Understand your people
To help you understand your people, listening tools need to identify relevant conversations related to your interests. In order to do that, tools need your instructions translated into their language: keywords. As soon as you have a topic in mind, you will type associated keywords into the filter tab of the tool. The more specific your query is, the more relevant the selection will be.
For instance, let’s say you own an Irish restaurant and you want to find out what people’s opinion about Irish burgers are. In case you just search for “Irish burger”, you will probably get reviews about Irish burgers – which is what you are looking for – but you will also get recipes or even comments about an Irish fisherman called Mr. Burger – which is out of your scope. As a consequence, you may want to add exclusion keywords, to narrow down your results as much as possible, such as “recipe” or “fisherman” in this example.
You now have an idea of the amount of mentions talking about Irish burgers. You can repeat the process for French, Dutch and other local burgers to compare volumes of mentions and deduce your “share of voice”, the percentage of mentions about Irish burgers. However, people might be talking a lot, but are they sharing positive or negative opinions? To answer these questions, most of the social listening tools on the market can automatically analyze “sentiment” in a text.
Thanks to integrated sentiment analysis, the tool will categorize mentions according to their “positivity”. Very often, a mention is either positive, neutral or negative. Adding this second dimension to your results will allow you to better understand whether more people shared a good or a bad opinion about Irish burgers.
- Human after all
Despite their powerful capabilities, it is important to keep in mind that social listening tools are based on recent methods and when it comes to semantic analysis, reliability will hardly reach 100%. The deeper you want to dive into this data, the less accurate results will get. For instance, sentiment analysis techniques are only efficient to a certain extent and are limited by the human language complexity. Indeed, ambiguous comments are often difficult to identify as positive or negative. Even the most popular social listening tools admit to get a 70% accuracy in sentiment analysis, not even in every language.
Nonetheless, the world of social media and big data is growing quickly, so are promising advances on text mining and machine learning. In 2014, a group of research from Stanford University claimed to have reached 85% accuracy in sentiment analysis. In a near future, as sentiment and semantic analysis are becoming increasingly accurate, we can imagine identifying not only sentiment variations, but specific emotions in mentions. We can imagine tools detecting purchase intent and thus identifying hot leads. With social media analytics, sky is the limit!