Machine Learning Can Reveal the Why Behind the Answer in Social Media Analysis

Remember math class in elementary and junior high school? (Sorry to dredge up particularly painful memories. It will be over in a moment.) When you took a test, your teachers probably insisted you “show your work.” They didn’t want you to just dash off an answer. They wanted you to demonstrate how you arrived at your answer — in detail.

“Showing your work” was annoying, yes. Especially if your handwriting made chicken scratches look like fine calligraphy. But the practice accomplished two important things. First, it prevented cheating; if you could document your logic process, cite your formulas (formulae?), and display your calculations, chances are you didn’t copy your answers from your neighbor. Second, in the unlikely case you entered a wrong answer, laying out the process could show a teacher where you tripped up. It’s a tool for improvement.

Knowing how one gets an answer is just as important today, especially when analyzing data to inform business decisions that will affect how large amounts of money get spent and made.  For instance, if your company wants to determine top-performing Twitter hashtags, keywords, and messaging sentiments for media planning and competitive analysis purposes, you should demand this same kind of transparency. Without it, you can’t trust that the recommendations are well-reasoned and accurate. Nor can you identify ways to improve the results you receive.

Fortunately, machine learning-enabled social analysis tools are available to help.  When evaluating a platform, here are key points to consider:

1. What’s the source of the information? What social media service is being analyzed? How deep does the analysis go? Twitter is currently the largest, most heavily used social media platform for topical discussions. If a service isn’t analyzing the full depth and breadth of Twitter, it might as well by watching MySpace.
2. How many tweets are being sampled? If you study statistics, you know that you don’t have to account for every response to get a clear, accurate picture of an event. But, at the same time, a sample size that is too small — like your company’s followers for example — can produce misleading results. Your sample size needs to be large enough to do the job right, and with the sheer volume of daily tweets, a trustworthy sample size is going to be a large number. For example, when doing social media analysis Northern Light takes all the tweets associated with a hashtag or author.
3. Are the tweets filtered by industry? Language and word meaning can vary wildly between its use amongst the general public and industry specialists, or even between industries. (For example, the term “crib sheet” is a decades-old slang term for a paper containing test answers, but to someone in the textile industry, it means a cotton sheet used to line a baby’s crib.) For a social media keyword analysis to have any value, it must have context.
4. Is the analysis based on the right metrics? Simple co-occurrence is the metric most often used by social media analysis tools.  But it can be very misleading because co-occurrence isn’t weighted by impressions.   Impressions are the sum of posts multiplied by the number of followers of each tweeter.  Analysis without impressions often gives just flat wrong answers.   With impressions, the light bulb goes on and the social media analyst knows what is going on.
5. Crucially, does the analysis consider the context of the social media conversation? Is the conversation about #telemedicine the same as the conversation about #telehealth?  Is the conversation about #financialplanning the same as the conversation about #retirementplanning?  Semantic analysis of the tweet content associated with each can answer such questions quantitatitvely.

Social Analytics from Northern Light has become a leader in machine learning-based social media reporting and analysis due, in large part, to the depth and detail in its cross-analysis and reporting capabilities. It focuses on the entire Twitter community to intelligently filter your search queries by industry, and cross-analyze across the criteria you choose, such as account, hashtag, keyword, location, or sentiment. It also detects semantic similarities between hashtags, identifies leading accounts for each topic, and displays all results in a precise percentage-based format, so as to yield maximum user insights.  To get a chart of your results, you’re just one click away.

In short, a social media analytics platform is only as good as what you put into it. And what you get out of social media analytics is only going to be as good as the platform you use. For the best results, use the best platform: Social Analytics from North Light.  Learn more about Social Analytics or contact us to start a conversation.