Context in Semantic Search

Imagine the most delicious donuts on the planet. Imagine they are now yours to have. Instantly. By some miracle of technology all you need to do is wish it and they will appear for you to consume. That’d be amazing, right? Except, you wouldn’t want to activate this technology the moment you’re on your way to the gym. Nor would you want to activate it at five minutes to midnight before you take yourself off to sleep.

That is the power of context. It can add extra value to something relatively basic by making it available at just the right moment it is most wanted or needed. But is also has the power to render it worthless by presenting it when it is least wanted.

Companies pay marketers huge amounts of money to fine-tune marketing campaigns so that messages appear at the most desirable moment, in context. Search engine optimizers try to create content in ways that will allow search engines to correctly divine the context of a particular product or service and present them when most desired.

Context collapse presents a huge challenge to marketers and companies alike. It also challenges search engine optimizers.

Search engines require context to divine:

  • Identity
  • Purpose
  • Intent

These three then project a sense of contextual value to particular search queries. Contextual value, in search, is represented through context awareness which is the ability of a program to behave differently depending on its surrounding execution context.

This has great applicability when it comes to capturing user intent by correctly understanding context and also creating algorithmic applications or filters that spot and highlight fake news.

The real purpose of this article however is to highlight what happens when a search engine that is driven by logic rules designed to help it better understand context comes across content that contains insufficient clues to understand its context correctly.

This has direct impact on website content, marketing, brand messaging and branding. It affects the way you approach all website content and it will also affect website audits whether they are intended to look at conversions (i.e. sales), branding or SEO.

The Brain And Its False Memories

While search engines are nowhere near in intelligence to a human brain yet, the fundamental structure of creating content out of information discovered and then generating a template that acts as a memory to use the next time a similar piece of information or a search query is encountered is sufficiently similar for the simile to work. Even more so, perhaps, given that search engines have clear, programmed imperatives in their operation that, unlike a human brain, do not deviate from their operation because of differing emotional states.

Two distinct pieces of research provide the breakthroughs in understanding we need here. The first, from researchers at Columbia's Zuckerman Institute shows how the brain, employing an evolutionary energy-saving system, encodes associative memories using sparse data models and ends up with associative memory disorders as a result that help explain PTSD and false memory syndrome.

Kanizsa Triangle and the brain's ability to recognize patternsWhat happens in these cases is that insufficient data forces the brain to create associative links with similar but unrelated events which then ‘trigger’ a response that is unwarranted. A loud car noise, for instance, forcing a traumatized vet to take cover and experience acute anxiety is one vivid example.

The second piece of research comes from MIT’s Picower Institute for Learning and Memory where researchers discovered that for every localized memory we create the brain runs two codes. One allows us to remember and store the current memory, for instance, having breakfast at home. This creates the familiarity that guides our behavior in that setting. At the same time the brain runs a separate code researchers labelled an “event code” that creates a generative abstraction which guides our behavior when we have breakfast anywhere else on the planet.

The research breakthroughs are important because they provide material evidence that backs up the intuitive claim that in the absence of sufficient information about an event we do not get an information vacuum. The brain is designed to fill in the blanks and when it doesn’t have sufficient information to do so, it makes it up. On the positive side of this we can “see” what is not there, like in the example of the Kanisza Triangle where the brain perceives a white upright equilateral triangle where none is actually drawn.

The negative side of this mechanism is that the brain can conflate memories or link up events that have some similarities without them being the same. PTSD that is triggered by a loud noise for instance is the brain linking up a traumatic event for which it doesn’t have sufficient, detailed information with something similar. The outcome can often be debilitating.

And this, now, takes us to search.

Search Engines Can Be Fooled

Think that in its drive to process the information of the outside world and index it; a search engine is not unlike a human brain. One that is hyper-focused on classifying the information it encounters and generating abstractions that can help it divine context in scenarios it has not previously encountered. The concept is the same whether we are talking about generative type abstractions in computational operations or the emergence of semantic hierarchies using generative adversarial networks (GANs) to scene synthesis in pictures and videos.

Unlike a human brain a search engine can neither be distracted nor dial-down its operational intent. That means that when there is insufficient data presented to understand context it will synthesize it from similar scenarios. The results, just like in a human brain, can be weird.

An example of this can be seen in examples where AIs have been allowed to be creative like in this picture of familiar objects which actually aren’t.

AI Draws image of indescribable objects

Visual search is at its best when it is able to understand context in order to render meaning. Unfortunately that is also its Achilles Heel. Visual search relies heavily on machine learning and the backbone of its programming permeates semantic search applications. Throughout that ecosphere context and deductive reasoning continue to present specific difficulties because both rely on real world knowledge.

The problem with that is that the real world constantly evolves. An off the cuff reference about “the boy who lived” instantly translates into Harry Potter for millennials, but the same reference in a 1970s context can only mean David Vetter who was known as “the boy in the bubble”.

Semantic Density

All of this leads us to “Semantic Density” a term I have used repeatedly in SEO Help. While most SEO practitioners think this has to do with Latent Semantic Analysis and the comparison of the density of semantic vector clusters in documents indexed by information retrieval systems it is actually related to Code Theory.

In Code Theory semantic density refers to the conceptualization we apply to complexity as 'semantic density', which describes how meanings are condensed and interrelated within knowledge practices.

In the real-world humans use mental heuristics to condense everything, save mental processing power and arrive at decisions quickly. To do that we rely on real world knowledge that involves a mix of cultural learning, traditions, experiential and formal knowledge.

Because search engines cannot be made to work like that, the possibility of them misunderstanding the importance of factual information and contextual value increases in direct relation to the semantic paucity of the information they index. Enriched semantic density requires both active ontology-building efforts and taxonomies of content. And this requires thinking, planning and execution that is guided by a true sense of identity, purpose and goals. Goals describe intent which leads us directly to the simplicity of the opening of this article.

Where does this then leave us?

Marketing Basics

Content that does not feel an audience need is poor content. Marketing that does not address the audience directly, is poor marketing. Search engine optimization even at its most technical can no longer be separate from marketing, sales and branding. Nor can a business, truly, look at each of these elements as separate departments that can each be allowed to do their thing.

Values drive behavior because they are reflected in choices and decisions which lead to actions. Actions create impressions and generate perception. Perception is reality. You can see how all this is linked in the image below:

Data, Culture and Context in Marketing

The 4Cs of content creation are now critical skills to have:


Finally, after all this, the TL;DR version of the piece.

  • Be detailed. Really, really detailed in all your content.
  • Be consistent. Do not mix your marketing messages to jump on the next bandwagon to gain an easy win. You increase the chances of it backfiring.
  • Be clear. State what you do, why, for whom. Explain why it is what they need.

That’s it.