How do consumers really feel about your brand? What about your competitors? Or the latest trend?
Companies can leverage sentiment analysis tools to answer these questions and better understand how audiences feel about any topic. As part of our ongoing Elements of Social Media Analytics series, this post dives into what sentiment analysis is and how brands can use it today.
Why Sentiment Matters
On January 5, 2017, something strange happened on social media: mentions of Toyota surged dramatically. Here’s how it looked:
At first glance, you might assume this conversation spike was a great thing for the car brand. But once we dig into this spike in volume, we see that the comments about Toyota were overwhelmingly negative.
Now the full picture starts to come into focus. The surge in Toyota mentions was prompted by a tweet from Donald Trump chastising the brand for planning to build a new plant in Mexico.
Toyota Motor said will build a new plant in Baja, Mexico, to build Corolla cars for U.S. NO WAY! Build plant in U.S. or pay big border tax.
— Donald J. Trump (@realDonaldTrump) January 5, 2017
Without analyzing the sentiment surrounding this spike in conversation, Toyota might have missed the importance of the event. Knowing why people are posting is much more powerful than simply knowing how much they are posting. Only by analyzing the sentiment of the conversation can Toyota come to the right conclusions about the unexpected surge in conversation.
But what is sentiment analysis? Where did it come from and how does it work? In this post we’ll look at the history, uses, and future of sentiment analysis, especially as it related to social media analytics.
What is Sentiment Analysis?
Don’t be fooled by it’s official-sounding name; sentiment analysis is a very fundamental, and very old principle. Like Toyota in the example outlined above, all brands need to understand how their audience feels about certain topics. Like what motivates consumers to buy Beats headphones instead of Bose? This is where sentiment analysis is instrumental.
In the past, when brands wanted to understand how consumers felt about something, they had to pay big money for market research firms to run focus groups and administer surveys. Today, sentiment analysis tools allows brands to answer these questions much more quickly and easily than in the past.
On the most basic level, sentiment analysis, or “opinion mining,” strives to answer the questions “what do people think?” or “how do people feel?” about a particular topic. To get more technical, sentiment analysis is the measurement of positive language and negative language.
“Sentiment analysis is the measurement of positive language and negative language”
While there have been studies about what people think and feel for decades, the term “sentiment analysis” was coined by researchers Bo Pang and Lillian Lee. Their work focused on the exponential growth of internet, and its potential as a source of data to analyze how people feel about anything.
In our modern lives, the largest publicly available collection of conversations exists online in the form of social media, presenting a huge opportunity for brands to turn this data into business insights. The endless stream of social media conversation allows the use of sentiment analysis of any topic from politics to Game of Thrones.
The most accurate way to measure sentiment is to have a human determine the sentiment manually. But new technology that uses natural language processing (NLP) allows us to apply sentiment analysis to text-based conversations on a large scale, quickly and accurately.
Whether you’re looking at the social conversation about a specific topic, brand, product, etc, sentiment analysis tools can automatically categorize posts based on the feelings they express. Today’s algorithm-based sentiment analysis tools can determine the feeling behind massive quantities of social media posts almost as accurately as a human, and even more consistently and at a much, much greater speed and volume.
Positive, Negative, Neutral and Beyond
The simplest sentiment breakdown would categorize posts as positive, negative, or neutral. This type of analysis is useful for looking at the big picture of consumer feelings on a topic, but obviously, misses some of the nuance of human emotion.
The next level of sentiment analysis categorizes posts based on specific emotions. But which emotions?
Some of the most agreed upon “universal emotions” were determined by psychologist Paul Ekman in the 1960s. Ekman’s six basic emotions are: Anger, Fear, Disgust, Joy, Surprise, and Sadness. Using these emotions gives sentiment analysis a deeper context for understanding the consumer feelings about a topic. It’s one thing to know that an advertising campaign was negatively received, but knowing that it made people angry instead of sad or disgusted can help you develop a plan to bounce back faster.
What’s the next step beyond a predefined set of emotions?
Custom categories that you determine depending on what you’re analyzing. Sentiment analysis tools that use machine learning allow you to create your own categories for researching consumer opinion on any topic. For example, if your company sells coffee, you could create custom categories specific to an analysis of how whiskey drinkers prefer to make their drinks. Example custom categories could include: neat, on the rocks, shots, craft cocktail (old fashioned, sazerac, etc), or highball (whiskey coke/ginger).
Use Cases for Sentiment Analysis
Sentiment and emotion add important context to social media analysis. Often, simple measurements of social media conversations provide the “what” without the “why.” What really matters is why more people are Tweeting about your brand or your competitor. Sentiment analysis provides a window into consumer feelings about any topic, brand, or product.
Identify a potential crisis
Without looking at sentiment, simply measuring your volume of brand mentions could be misleading. You might assume an increase in mentions is a positive sign that more people are talking about your brand. All press is good press, right? But what if all those mentions were negative?
As we saw in the Toyota example, a spike in conversation about a brand isn’t always a good thing. Sentiment analysis helps brands recognize events that need to be addressed to avoid the blossoming of a full-on crisis.
Once it was clear that the spike in conversation about their brand correlated to a surge in negative sentiment, they probably recognized they had to take action, and they did. They released a statement explaining the situation.
Evaluate overall brand health
The Toyota-Trump exchange is an example of a one-off event that had a large impact on audience sentiment, but how did it fit into the company’s overall relationship with its audience? That’s where overall brand health comes in.
There are a lot of metrics to look at when measuring your brand health, but sentiment is key. Establishing a baseline for how people feel about your brand and products is essential to tracking the performance of your brand over time. For Toyota to identify the spike in negative emotion, they need to know what their net sentiment looks like normally. Is positive sentiment on the rise, the decline, or totally variable day to day? What factors (ads, product features, press, etc) are influencing swings in sentiment?
Track the reaction to new products or branding
Knowing how people feel about your new products or new branding is important. Think about Apple’s release of the iPhone 7. A sentiment analysis could have helped Apple pinpoint the removal of the headphone jack on the phone as the main driver of negative sentiment in the conversation.
Get social insights delivered to your inbox.
Beyond analyzing your own brand and your competitors, there are plenty of ways to use sentiment analysis for broader market research. For example, how do consumers feel about self-driving cars and new developments in autonomous vehicles and technology?
Emotion analysis of the self-driving car conversation back in April 2016 provides a clear answer: fearful.
As an automaker, it would be important to identify and understand the big spike in fear in mid-April. A closer analysis shows that the spike in conversation volume stemmed from a viral Tumblr post questioning how autonomous cars would react in an emergency.
Insights like this can help automakers understand how the public feels about new car technology. In this case, it’s clear that car companies have work to do convincing the public that driverless cars are safe.
To learn more about how to use sentiment analysis to reach actionable business insights, check out our Business Insights From Social Media Data guide.