Many audience building strategies rely on bucketing people based on their similarities. However, we find that marketers create powerful, high-performing segments when they consider potential customers’ unique content engagement patterns -- and the significant differences -- to turn interests into actions.
As mentioned before, with Semantic Behavioral Targeting marketers build segments based on the words and content consumers engage with -- giving them the power to find their “impossible audience.” We’ve discussed how Semasio uses natural language processing as an information-preserving method to extract important information from content online. In this post, we delve deeper into this method to discuss the basic methodology we use to gauge semantic significance, why it works, and why it’s the right approach for audience targeting.
The Search for Significance
Measuring semantic significance is simple in nature, but powerful in practice. The approach is as follows: for a single page, first eliminate all non-editorial content. Then, compare the likelihood of each word in that string to appear, given its likelihood to appear in the language in general. This part bears further explanation: By preserving and analyzing editorial content across the internet, we build a statistical model of the likelihood of a word to appear in a given context. The word democrat in a phrase like I am a____ is likely to appear, relative to the rest of English lexicon. A word like astronaut, significantly less likely. In this way, we measure the semantic significance of a word: the less likely a word is to appear on a webpage, multiplied by the frequency with which it appears, makes the word more significant. This is an extremely efficient way of distilling the full richness of information on page into an informative, and scalable profile.
The Significance of Significance
By extracting the significant terms on each page, we wade through the less useful information – the stop words, the pronouns, the words that are so rampant on the internet that they tell us nothing about the unique content of the page (For example, try finding a news article on the front page of The New York Times in 2018 that does not make at least some reference to Donald Trump.) – to figure out exactly what a user is consuming on that page. This approach is smart, in the sense that our model of a given language always learns. As a word becomes more common, as a word like sandy did around the time of the hurricane, its weight in the model decreases; as it becomes less common, it increases. As the internet grows, and our everyday language itself changes, so does the model.
The approach is also flexible in that it adapts to any language for which individual words can be extracted from the written content. Every day, clients use our system to process languages with Roman alphabets, like English and Spanish, languages with special characters like German and Danish, and non-roman languages like Arabic.
The Audience Application of Significance
Why should a marketer care about semantic significance? In another article, I outlined why language helps you learn about your audience. Firstly, it’s simply a precise, scalable way of assigning a unique profile to every single user in the web: we build user profiles by dynamically appending the significant terms across the pages they browse to their profile. So, as long as two users have not visited the exact same web pages every day of the past 60 days of their lives, they will never have identical profiles. Semantic significance makes users unique. It also avoids the opposite issue of giving too much detail for a user – more than what could be relevant to a marketer.
By distilling all the pages a user searches into a few, transparent terms, semantic significance allows us to get the relevant information about the user: what are the topics this user cares about? Once you understand the unique motivations of your potential customers, it’s much easier to turn those significant interests into attributable actions.