With social listening tools now commonplace, plenty of companies keep track of what people say about them online. What they find, however, is there’s a big gap between tallying brand mentions and deeply understanding what customers (or potential customers) really think and feel.
Most social listening tools can’t make sense of subtlety: They can’t make out complex human interactions, such as expressions of irony or playfulness. They still struggle to interpret still and moving pictures. They can’t keep up with fast-changing abbreviations and memes.
Technologists are working to improve sentiment analysis, which helps brands truly understand what their customers think. Somewhat.
In the last few years, we have gotten closer to understanding intent. Rather than simply saying someone is positive or negative about our company, we can train machines to recognize more nuanced emotional states (e.g., sad, irritated, angry, elated) and apply these sentiment assessments to specific brands and products.
Yet serious limitations still exist. More established techniques still rely heavily on our giving the machine a “dictionary.” Plus, some techniques do not scale well to big data volumes. Perhaps most frustrating is that to get meaning right, machines must read “well-behaved” text – copy that is spelled correctly, uses standard grammar, and avoids slang.
The next frontier for sentiment analysis will integrate artificial intelligence to create machines undaunted by such complexity, and able to keep pace with the subtle and rapid evolution of social chatter by teaching themselves rather than relying on human decoding.
I spoke to Seth Grimes, an analytics consultant with a particular interest in human data and founder of the Sentiment Analysis Symposium, about the state of text and sentiment analysis and what it will mean for marketers.
Why should brand marketers care about topics so much in the realm of data science?
Emotions drive actions. Sentiment analysis is about quantifying emotions – including (and especially) unsolicited feelings expressed on social media. The number of ways that customers and prospects interact with companies has increased significantly over the last 10 years. So, as always, the first Big Data challenge is to filter. That means selecting not only the platforms and sources that are of greatest importance, but also the information that’s of greatest relevance.
How should marketers get started?
We start by figuring out what data are important to the organization, whether we’re talking textual data or any other type. What drives your business volume, profitability, and other measures?
What really interests me, and what I think should interest marketers, is what I’ll call signals – one of which is intent. Intent is critical because it can predict action. For example, “Is this person shopping to buy a product like my product?” “Is this person unhappy and needing some form of attention?” “Is this person about to return the product for a reason that is addressable?”
Sentiment is one ingredient of intent. If someone is happy, sad, angry … that can be determined via sentiment analysis technologies.
How refined is our ability to measure sentiment and intent?
There are a lot of different technologies out there – some of them too crude to be accurate or useful. For instance, when using text-analysis technologies, you may start with a simple word list (i.e., a topical vocabulary). “Good” is a positive thing and “bad” is a negative thing. But don’t stop there. You need at least a bit of language understanding, for instance the ability to handle a negator when “not good” is bad.
Many tools struggle with context. An example I hear over and over again is “thin” – good when you’re talking about electronics, but bad if you’re talking about hotel walls or the feel of hotel sheets. To do sentiment analysis correctly, you need refinement. You need customization for particular industries and business functions.
The market, unfortunately, is polluted with tools that claim to have sentiment abilities, but are too crude to be usable. Even with refinement (e.g., the ability to handle negators and contextual sentiment), approaches that deliver only positive and negative ratings don’t take you very far.
Is sentiment analysis strictly for mature marketing organizations?
There are definitely easy, inexpensive entry points that can meet basic, just-getting-started needs: tools for social listening, survey analysis, customer service (handling contact-center notes, for instance), customer experience (via analysis of online reviews and forums), automated email processing, and other needs. These technologies are user friendly, available on demand, as a service.
The real question is what the data will actually tell you. At the root of it all, you have to ask: Is the availability of information going to tell you anything that will help you run and optimize your business or your business processes?
Grimes offers these companies as ones to watch on the frontier for various technology-driven sentiment analysis.
Text mining: Facebook, Google, LinkedIn, Pinterest and the like have improved their bench strength in artificial intelligence and machine learning through key hires and acquisitions. While many advances are still academic (or hidden in these companies’ non-disclosed R&D labs), other companies are openly pursuing unsupervised learning for text analysis. They include Digital Reasoning, Luminoso and AlchemyAPI.
Image recognition and analysis: Image analysis now automatically identifies brand labels in pictures. A sampling of companies developing these technologies and the analytics to go with them include VisualGraph (now owned by Pinterest), Curalate, Piqora (nee Pinfluencer), and gazeMetrix.
Emotional analysis in images, audio, and video: These companies promote analysis of speech and facial expression primarily for structured studies (e.g., ad or media testing, point-of-sale surveys or monitoring, event feedback). But as the technology matures, it’s easy to imagine that marketers may use these companies to harvest the unsolicited, unstructured audio and video content of mobile and social data:
• Affectiva conducts webcam emotional analysis for media and ad research, including development tools to integrate emotional study in mobile apps.
• Emotient performs emotional analyses in retail environments, evaluating signage, displays, and customer service.
• EmoVu by Eyeris tests the engagement level of both short- and long-form video content.
• Beyond Verbal studies emotion based on a person’s voice in real time.
Looking for extra guidance on understanding and using content marketing technology? Get helpful tips in CMI’s eBook: Building the Perfect Content Marketing Mix: Internal Processes and Content Marketing Strategy Tactics.