From voice assistants to self-driving cars, machine learning is everywhere. Every day, a new product, app, or service announces that it uses “machine learning” to get better and smarter. But because the term can mean so many different things, it’s hard to understand exactly what machine learning means when it’s applied to a new technology.
As part of our Elements of Social Media Analytics series, this post covers how machine learning applies to social media analysis and how it can help brands uncover deeper consumer insights.
So does this mean that brands should rely solely on manual social media analysis? Of course not. While you could manually sort through a long list of posts to find answers to your questions, that process isn’t scaleable and can’t keep up with the speed of consumer opinion and trends. [We’re talking about trillions of social media posts after all.]
Enter machine learning. For social media analysis, a machine learning algorithm combines human understanding with the scale and speed automation. Analysis built on machine learning allows you or your analysis team to train the algorithm to categorize posts just a human would.
Imagine you’re trying to analyze the social conversation about Tide detergent. Using a keyword-based search for “tide” would return results for people talking about the ocean, “turning the tide,” and the Alabama Crimson Tide in addition to any posts about the detergent.
Machine learning-powered social analysis enables you to pinpoint exactly what you’re looking for (here, the brand Tide) by using example posts to train the algorithm to recognize the patterns of language that indicate that a post is about Tide the brand, instead of something else.
within the first 20 minutes of wearing my white jeans i have acquired a massive stain. where my tide to go pen at
— joely (@joelyhand) February 21, 2017
And that’s just the start. Machine learning’s ability to interpret the nuances of language and return the most relevant results reaches far beyond identifying brand mentions. Social analysis using machine learning makes it possible to answer questions like:
- What drives people to purchase my product?
- How are customers using my product?
- What consumer trends provide the biggest business opportunities?
What is Machine Learning? (for social media analysis)
Arthur Samuel, an early pioneer in the field of artificial intelligence, defined machine learning as “the subfield of computer science that gives computers the ability to learn without being explicitly programmed.” Additionally, Wikipedia describes machine learning as “a scientific discipline that explores the construction and study of algorithms that can learn from data.”
What does this mean for social media analytics? Machine learning gives an analysis tool the ability to learn exactly what you’re looking for in social media posts, and categorize posts based on that training.
But why is this necessary? Why isn’t there an algorithm that doesn’t need to learn, it just already understands what I want it to find? The reason lies in the unstructured nature of social media data.
Structured data vs. unstructured data
Structured data refers to the kind of data that is organized and displayed in a database with rows and columns, making it straightforward to work with. Examples of structured data include everything from social shares and likes to sales numbers and business addresses. These types of data are easy to sort and categorize.
Unstructured data is variable and complex, making it much more difficult to sort, categorize and analyze. Examples of unstructured data include emails, images, and any form of human language in a conversational format (like social media posts).
Simple tools can easily analyze the structured, volume-focused parts of social media such as how many followers your brand pages have, how many shares your last post got compared to other posts. But it’s the unstructured social data, in the form of conversations and images, that holds the answers to the more important business questions. That unstructured data requires a deeper type of analysis.
Rules- vs. pattern-based social media analysis
When working with unstructured data, it’s crucial that your method of analysis has a high degree of flexibility to conform to what you’re looking for. While simple rules-based analysis works fine for making sense of structured data, getting insights from unstructured data requires pattern-based analysis.
A rules-based analysis would show you all posts with a specific word (eg. “Tide” AND “detergent” NOT “ocean”). While this could work, to show you posts about your brand, it doesn’t help you uncover much in the way of business insights.
Get social insights delivered to your inbox.
Unlike rules- and keyword-based analysis, machine learning analysis can classify posts based on specific examples that you choose. Through a machine learning algorithm, a computer can be trained by a human to recognize patterns in posts that match the training posts. This opens up many more possibilities for customized analysis categories.
For example, machine learning could identify posts that fit into categories like “happy Tide customer,” “unhappy Tide customer,” “planning to buy Tide,” “evaluating detergent options,” or any other way that Tide might choose to classify posts about their brand.
No fully automated tool will be able to define these customer lifecycle categories for you, but hand-coding thousands of posts wouldn’t be an efficient use of time. This is where machine learning makes a big difference. With machine learning, computers use training data to accurately classify posts just like a human would.
How Machine Learning Enhances Social Media Analytics
There are countless applications for machine learning technology in social analysis, including improvements in sentiment analysis, audience analysis, image analysis and more. Here are some specific use cases:
Analyze the buying cycle for your product
Machine learning opens up the ability analyze previously unmeasurable aspects of consumer behavior. With the right training data, you can categorize posts by their stage in the buying cycle.
Answer complex questions
Machine learning also opens up the ability to answer complex business questions. Following the Roku/Apple TV example, both of those brands would probably be interested in answering “Why do people cut the cord/get rid of cable?” Creating custom categories for the most common reasons and training the algorithm to classify them will allow you to get detailed data on why people are making this decision. Previously, this type of consumer option data would only have been available through surveys or market research.
Detailed analysis of any language
Because machine learning relies on training data, you can train it to analyze and categorize posts in literally any language (Klingon, Elvish, and Dothraki included). All that matters is that the person providing the training data knows the language well enough to provide appropriate examples posts for the categories you want.
Filter out spam
Social media analysis often gets derailed by spam posts. Image you’re Apple and want to see how people are talking about the iPhone, but half of the posts are spam about winning a free iPhone 7. Machine learning ensures that the results of your analysis only include the relevant posts based on your training.
Get the most accurate results
Machine learning inherently increases the accuracy of social media analysis. While manual human analysis is still the most accurate, machine learning comes much closer than than traditional rules based analysis.
For more detailed information on how to structure your machine leaning-based social media analysis, download this free guide: Business Insights from Social Media Data