Social Listening: When #AI Takes Over

A Guest Post by Albane Flamant, Marketing Manager – US, Talkwalker

Once upon a time, brands and agencies relied on human analysts to identify brand mentions and extract insights on their campaigns. Today, because of the sheer online data volume, this approach is no longer sustainable: in the next ten minutes, there will have been over 3 million new tweets posted online, along with hundreds of thousands of Tumblr articles, Instagram posts, YouTube videos and much more. Let’s not even start talking about more traditional online sources like blogs, news sites, and forums.

In order to make sense of all that “noise”, social listening, monitoring and analytics tools have become essential parts of the marketing and PR toolbox. Here are three ways artificial intelligence is helping unlock new data insights within these platforms.

Image & Video Recognition – Analyzing visual content

Cisco predicts that by 2021, 80% of online content will be videos. We’re not yet at that stage, but the rise of visual content is undeniable, with Instagram reaching the milestone of one billion users (and counting) in June 2018.

Because of this new reality, brands can no longer afford to just monitor text mentions – they also want to know what happens in all of these images and videos. With the help of machine learning, social listening platforms are now able to analyze visual content to identify logos, scenes and objects.

The idea is not only to detect hidden brand mentions (visual posts where the brand is not mentioned explicitly in the companion text), but also to understand the context in which your brand, products and services are being discussed. Use cases go from brand monitoring to event sponsorship ROI, user-generated content identification and product research.

Example of video recognition technology
on the Talkwalker listening & analytics platform

How does it work: algorithms are trained to recognize patterns in images and videos by going through huge data sets (this process is called machine learning). 

Sentiment Analysis – Putting tweets in context

Sentiment analysis technology used to have a bad rep in the PR industry because guess what? Platforms had trouble understanding all the small nuances of language! This is mainly because the “old” approach was to look at keywords to determine sentiment (whether the post is negative, positive or neutral).

The first reflex of brands and agencies was to turn to human analysts to manually sort through the data and correct sentiment. Yet there were too many inconveniences: sentiment was still subjective based on the analyst, and at the end of the day, there was just too much data to sort through to be able to rely on their work.

Imagine: a brand like Starbucks averages over 63,000 daily mentions! How do you keep up with that? By the time you’d detected a surge in negative sentiment, the crisis would already be well underway. 

With the help of machine learning, platforms are now able to look at the full context of a sentence to determine sentiment and start to understand basic irony (again, the trick is to train them with a big enough data set).

Brands and agencies also have the option on specific social listening platforms to train their own custom models, so the platform can adjust to the sensitivities of each industry and even understand which brand is in front of the screen.

Case and point: how would you classify the following tweet if you didn’t know if you were doing it from the point of view of Pepsi or Coca-Cola?

Sure, I like Coke Zero. As in zero cokes. This is Pepsi country.

It’s been a pleasure doing business with you.

— Cauliflower Shane (@KingDomeGnome) April 9, 2019

Custom models – Filtering out the noise

As previously mentioned in this article, custom models help brands and agencies get the relevant results to surface. Let’s imagine you’re a brand marketer for Apple. How do you filter out all the fruits, Big Apple, apple martinis and other irrelevant junk from your queries.

The answer used to be through sophisticated boolean queries. The problem, however, was that these queries were complicated to write for the average marketing or PR professional, and often included or restricted too many results.

Enter custom models, which prompts you to tag results as relevant or irrelevant without any data training. Once you’ve provided the tool with a big enough training set, it applies your model to any future results. Too complicated? Here’s a quick video to sum it up.

AI: Friend or foe?

Artificial intelligence is still perceived by many in the PR & marketing industry as a danger to their job. I’d argue the reality is much different. AI helps us get rid of manual tasks, it allows us to automate processes so we can focus on the more creative aspects of our jobs.

Albane Flamant, Marketing Manager, US, Talkwalker

Albane is a marketing manager at Talkwalker, where she coordinates the brand presence’s in the United States, from influencer relationships to content marketing. She lives on Twitter and works with social media experts from all over the world on white papers and webinars about new technologies and digital trends. She always has at least one book in her purse and has lived in 6 different countries over the last 12 years. Feel free to reach out to her on Twitter (@AlbaneFlamant).

Featured Image Photo by Christopher Burns on Unsplash