What’s Google’s End Game for Machine Learning and Semantic Search?
The past few years have been exciting for search marketers (or terrifying, depending on how much you tolerate change). Google has introduced a bevy of new technologies and systems to improve its already-impressive search algorithm, and competitors like Bing, Apple, Microsoft, and even Facebook have introduced or improved competing systems to make search even easier for consumers.
At the forefront of these improvements has been a focus on one of the hardest problems in computer and artificial intelligence history: the acquisition and understanding of natural language. While machines excel at programmatic and logical tasks (like solving math equations or identifying objective data patterns), they have trouble picking up on subtle distinctions in language that only a native speaker can intuitively comprehend. For example, to us, the phrases “Where’s the nearest burger joint?” and “I want to eat a burger” convey similar intentions, but to a basic machine, these are very different requests. The first is a request for information, independent from a desire. The second is a desire, independent from a request for information.
Google has dedicated the past several years to breaking down the language barriers and learning difficulties that its machine algorithms have faced, but what’s the end game? Is there a point where Google hopes to fully understand the intentions and linguistics of the human mind, or is this just one of many paths of development to come?
Semantic Search Roots and Hummingbird
Semantic search first came on the scene in 2013, with the introduction of the Hummingbird update. Prior to that, Google used keyword clues to map queries to existing pages online — as an example, in our burger scenario above, Google might have looked at the word “burger” and find pages online with frequent mentions of that word. There was no understanding of user intent, or that those queries might differ by seeking a restaurant burger versus a home-cooked burger. Hummingbird introduced an intention-based distinction; though not perfect, it transformed the process of the algorithm to factor in user desires and “understand” what pages are about, rather than just what keywords they contain.
Personal Digital Assistants and Voice Search
Personal digital assistants, including Google Now and contemporaries like Siri, demand a higher degree of semantic understanding. There are several new hurdles here, including translating spoken words to input text, finding the appropriate type of search, and responding in an intelligible way.
Artificial intelligence algorithms have iteratively progressed these technologies to the impressive level we see them at today, though the core of search results are still fetched using Google’s standard search algorithm.
The obvious motivation is to make it easier for searchers to use, but the secondary motivation is the change in user habits it forces: voice search demands the use of conversational inputs and contextual clues, which allow for stronger, more relevant results than traditional keyword-based inputs that most typists would rely on for brevity.
The Emergence of RankBrain
The big news on the semantic front this year was the introduction of RankBrain, a machine learning algorithm that works in conjunction with Hummingbird. For the sake of conciseness I’ll summarize its intention: to help Google understand bulky, complicated, or otherwise ambiguous conversational user queries. Think of it as a translator between poorly thought-out user inputs and queries that the algorithm can logically address. And because it capitalizes on machine learning, it’s capable of updating itself, rather than relying on the manual tweaks and pushes of Google developers.
A Rising Trend in Related Questions
To add a level of intrigue, Google is increasing the prevalence of rich answers (the concise “answers” to your questions that occasionally pop up above your traditional search results), and even more recently, the prevalence of “related questions,” which encourage users to explore their topics further. What’s especially interesting is that the answers to related questions currently differ from their respective answers when presented as a rich answer, implying the two services are working from separate sections of Google’s algorithm (presumably, the Knowledge Graph and RankBrain). In any case, Google appears to be ramping up their ability to not only understand user questions, but answer them concisely.
Three Major Predictions
To extrapolate the meaning from these otherwise objective observations, I want to offer three predictions about how Google will choose to develop over the course of the next decade:
1. Machine learning will become the new goal. Machine learning only exists in one niche corner of Google’s search algorithm; I imagine it will start to spread to other areas, including content quality analysis and backlink context evaluation.
2. Google will try to reduce every query to the form of an answerable question. With the rise of rich answers and related questions it’s clear that Google likes providing users with direct information. I imagine we’ll see a greater effort to push these answers forward, including the transformation of basic queries into more complex, answerable questions.
3. New tech will push more people toward voice search. Voice search naturally lends itself to complex semantic questions and direct-response answers. More people using it means more data for Google, better results for searchers, and an overall more harmonious system. Accordingly, I imagine Google will start pushing more people toward voice search however they can in subsequent years.
Use these three predictions however you see fit. You can start adopting new content marketing strategies to cater to user questions, hedge your SEO bets with new tactics that protect against the frequent changes machine learning could bring, or just sit back and wait for an easier, more intuitive means of search for your own personal use. There’s no guarantee what the future holds, but looking back at how far we’ve come in just the past few years, I imagine the next decade to be a groundbreaking one.