First, let’s agree that social listening should be standard operating procedure for any brand or company by now, whether you do it using Google Alerts and saved searches or through tools like Netbase, Sysomos or Radian6.
Let’s also agree that the sentiment expressed toward brands online are an important aspect of social listening. It’s not enough to simply monitor mentions of a brand or company; you also need to know what people are saying about it and how they feel about it.
[Tweet “The Problem With Sentiment Analysis #SocialMediaMarketing”]
The Limitations Of Sentiment Analysis
Let us further agree that to date, sentiment analysis has been woefully lacking. You’ve likely heard explanations for why there is a problem with sentiment analysis.
The problem is keyword-based sentiment detection can’t understand situations like this:
“Oh, yeah, Fast Food Restaurant. I just LOVE the 30 minute wait for my food.”
We humans understand sarcasm. We understand the sentiment of this comment is clearly negative. Yet a machine would flag it as positive, possibly even very positive because of the all-caps LOVE. How terribly wrong.
The solution, as Penn goes on to show how to do, is for a human to perform sentiment analysis on a sample dataset. But that is extremely labor-intensive and thus, does not scale.
Twitter’s advanced search provides an option to check for sentiment in a very simple way by checking a box for a smiley face or a frown emoticon. Predictably, that will yield such false positives as this tweet, which is supposed to be a positive tweet about Donald Trump:
Goodmorning everyone 🙂 …. except Donald Trump. Time to get ready for work.
— K. (@kswaggston) May 26, 2016
Or, conversely, this one, which is supposed to be a positive tweet about Hillary Clinton:
Hillary Clinton is saying that this last report on her by the Inspector General will not affect her run for President. LOL Delusional, 🙂
— Ron Surrett (@Tsali13) May 26, 2016
That’s a sad, sad state of affairs, isn’t it? Kinda makes you want to cry.
Text analysis, especially when there is precious little additional text to place it in context such as is the case with a 140 character tweet, will always prove lacking unless it is coupled with additional sets of data.
…social analytics firms have moved to supplement sentiment analysis with other metrics…” by providing “some sort of demographic information about who is posting about a topic or help pinpoint who is influencing an online conversation. Paying attention to ‘who’ rather than just ‘what’ can help these firm’s natural language processing engines interpret language—knowing a user is 14 rather than 40, for instance, might help interpret her use of the word ‘killer’—but it also provides their clients with more content around that analysis.”
Sentiment Analysis Needs Greater Context
There’s the rub. The “context” is not necessarily the what but the who.
The addition of basic demographic information about who is expressing the sentiment will no doubt help improve accuracy but even that is lacking. What we need, is extremely rich user data. And when you think of rich user data, what is the first word that springs to mind?
Facebook, of course!
Facebook catalogs everything you do on the site (and even things you do off the site with tracking pixels) and stuffs it all into the Database Of You. Think of all the things you tell Facebook through your use of the site:
- Your name and gender,
- Your email address, physical address and telephone number,
- Your martial stats,
- Whether or not you have kids,
- Your employment history,
- Your educational level,
- Who your family, friends and co-workers are,
- Who you went to school with,
- What brands you like,
- What kind of entertainment you like,
- The news you read,
- The groups you’ve joined,
- The content you’ve shared, commented upon, and liked.
Facebook recently changed the dynamic of the lowly Like button with the introduction of emoji-like Reactions, which include a Love, Hahah, Wow, Sad and Angry button.
Marketers have long discounted the value of a Like because of its inherent ambiguity and that’s precisely why Facebook added Reactions. Clicking Like is not an appropriate response to someone who has announced the death of a loved one, for example.
Facebook Reactions offer users a way to more precisely express their emotions while giving Facebook much more accurate sentiment signals.
The social network could also take it a step further and eventually use this emotional data for ad targeting, which would allow advertisers to zero in on users based on the specific reactions they had to a type of content, said David Erickson, vice president of online marketing for public relations firm Karwoski & Courage.
“Conservative political campaigns, for instance, could target Facebook users who reacted angrily to articles about Hillary Clinton or Bernie Sanders or President Obama and vice versa for liberal candidates,” he added.
Think about that.
Facebook now has the most complete context within which to analyze sentiment.
The best sentiment analysis tools rely on their access to Twitter’s firehose of data but have only limited access to Facebook data. That’s great, but Twitter has nowhere near the rich user data to provide context about the who behind sentiment which Facebook has.
If Facebook makes some of that data available to Page owners, marketers will have more precise metrics by which to judge how people react to their own content. Reaction analytics will provide greater insight into what types of content provoke specific emotions. That understanding can then be fed back into future content creation efforts and, much to Facebook’s delight, social advertising campaigns.
The Evolution Of Sentiment Analysis
Sentiment analysis will continue to evolve as the technology for performing the analysis evolves.
Google’s investments in machine learning and artificial intelligence are paying off to the point where the company has open-sourced some of its text analysis technology, called SyntaxNet and, amusingly, Parsey McParseface. Dieter Bohn of The Verge reports:
SyntaxNet is the overall framework for parsing sentences, called a ‘syntactic parser.’ Parsey McParseface is the English language plug-in for SyntaxNet. Google claims that it can correctly identify the subjects, objects, verbs, and other grammatical building blocks of sentences as well (or, in some cases, better) as trained human linguists — achieving 94 percent accuracy on English-language news articles.”
The technology for handling, managing, and analyzing massive amounts of data has advanced so rapidly that an entire industry has been created to satisfy the demands for processing “Big Data,” as this chart showing search volume growth for the topic of Big Data illustrates:
Accurate sentiment at scale is in your future.
To quote cyberpunk novelist William Gibson, “The future is already here — it’s just not very evenly distributed.”
And at this point, the future looks to be distributed almost entirely to Facebook.
Get The Success @ Creative PR Newsletter
The mission of the Success @ Creative PR newsletter is to help you succeed in your communications career by giving you valuable tips, trends, cool tools, insights and inspiration that will help you throughout your career. Get new marketing stats every week! Click the button below to subscribe: