This post is the second in a series on the eContext API and how our clients get value out of semantic classification. To learn more about our technology and its uses, feel free to get in touch.
In our last post, we presented semantic classification as a force for organization. By classifying both search keywords and ecommerce products to a shared taxonomy of concepts, businesses can improve the shopping experience by matching searches to results, regardless of ambiguities in word choice on either side of the equation.
This organizational value of eContext’s classification extends beyond ecommerce, improving user experience in areas such as digital publishing and enterprise document management. We will touch on a few of these applications in future posts.
However, organization is only one of the core benefits of effective classification. Today we’re going to focus on a separate (though related) advantage: the ability to learn from large volumes of data. Specifically, we will talk through some steps by which brands, agencies, and market researchers use eContext to study audience behavior.
The Value of Knowing Your Consumers
Broadly speaking, when brands and agencies gain a comprehensive understanding of their audiences–distinct interests, concerns, trends in behavior, etc–they become better equipped to maintain and grow those pool of users. That means more relevant messaging, products and services tailored to an audience’s needs, and marketing that’s cheaper for being more efficiently targeted.
To give a few specific examples, here are a few objectives eContext clients have worked towards by analysing their data with semantic classification:
- Improve advertising ROI by identifying websites and social channels with low competition among advertisers but high relevance to the target audience
- Increase brand recognition by developing a corporate identity that reflects customers’ values
- Craft consistent messaging, across social and marketing channels, that resonates with the distinctive interests of the target audience
Insights Everywhere — If You Can Find Them
So, to meet objectives like the ones listed above, what kind of information do you need? Typically, eContext clients utilize both user-generated content (social posts, blog entries, customer surveys, etc) as well as activity-describing data such as browsing information, website performance statistics, or purchase histories. (It should be noted that eContext does not provide this data ourselves; clients either have first-party access to this information, or license it from third-party data resellers.)
Latent in these sources is a quantifiable profile of what your target audience cares about, provided you can line up all those sources and interpret them in tandem. We’re talking about stores of information that are simply too large to manually process, but if you’re learning about audiences from their browsing and social habits, you need to understand the topics discussed within all that content. That’s where eContext comes in. Our API allows clients to label data, using several different parsing and classification functions, in a way that normalizes all that disparate content to a single hierarchy of topics.
Learning from Audience Data with the eContext API
Here are the steps most eContext clients follow when they obtain and classify data to learn about target audiences:
Step One: Gather Relevant Data
Clear objectives are a must when sourcing data for classification. Are you trying to maintain your current customer base or make inroads with a competitor’s? Are you trying to increase CTR in digital advertising? Reduce bounce rate once customers arrive on your site? Considerations like these will determine what kinds of data you should be focusing on.
For example, one eContext client sought to solidify the brand loyalty of their current customer base. The client attempted to meet this goal in two ways. First, they wanted to stack ad impressions across a typical customer’s browsing journey in order to make the brand more memorable. Next, the client wanted to establish a social presence that directly appealed to the target audience’s most distinctive interests (meaning the interests shared by the brand’s audience that were unusual relative to an average consumer).
To support these goals, the client obtained a few specific data streams:
- Browsing histories of the brand’s customers, including sites visited directly before interacting with the brand
- Performance statistics for individual articles in popular digital publications deemed relevant to the brand (“We know these sites are popular with our customers, but what specific content are they most interested in?”)
- Social media output from users who either provided their usernames directly to the brand or who mentioned the brand in recent social posts
If you’re interested in using high-volume data to learn about audiences, but aren’t sure which kinds of data you should be analysing, feel free to get in touch. We’ll be happy to consult with your on your objectives and give our advice on the best types of information to analyse from a semantic perspective.
Step Two: Classify It
Brands, agencies, and researchers have access to a whole world of overlapping technologies aimed at making sense of data. At this point, it’s worth stressing our belief that text data should really be understood semantically, ie, in terms of the topics mentioned. Sourcing or interpreting big data by syntax alone can lead to skewed results because language is inherently idiosyncratic. If a shortsighted data analyst says, “Let’s see how many of these Facebook posts talk about roller coasters”, he or she might end up with unreliable results due to the number of users who use the phrase “roller coaster” metaphorically.
eContext’s comprehensive semantic classification looks beyond word choice to identify the actual topics being discussed, labeling content using an organized structure to help clients view different data sources alongside one another for a holistic view of an audience.
After clients have sourced relevant data, they pass it through the eContext API using one or more of the following calls:
- Classify/keywords – for analyzing search queries and extremely short form-field responses
- Classify/social – for analyzing social media content, blog posts, open-ended survey answers, and other forms of UGC
- Classify/url – for analyzing websites (except those that are blocked by a paywall or other obfuscator)
Step Three: Interpret
At this stage, your data has now been structured to tell a coherent story about your audience. You’re no longer looking at a list of inscrutable URLs, for example, but a sequence of websites, labeled by topic, that describe a consumer’s’ various interests.
So the last stage of interpretation involves pulling all these different streams of information together, looking at them in terms of the standardized topic labels that have been applied to each, and identifying trends, features, or correlations that will support your initial goals.
- Compare the demonstrated interests of a target audience against a generalized control group in order to discover which interests are most unique to your desired customers
- Segment an audience by geographic information or other demographic indicators
- Correlate different topical interests to identify look-alike opportunities for organic brand growth
By combining traditional market research strategies with the structured, granular interest identification made possible by semantic classification, eContext helps brands and agencies understand their audiences to grow attention, loyalty, and ultimately revenue.