Narrowing things down one variable at a time is a great way to find that one ideal product in a lineup of dozens of others like it, each with a minor difference that might make it unique. Often, this is done via keywords and the logic behind it is simple: every keyword represents a variable that can be specified to narrow down your choices. Consider the following example:
Jane is looking for a beige, cashmere, crew-neck sweater that has rib-knit wrists, in a size medium. She visits her favorite clothing website and types “beige crewneck cashmere sweater” into the search box. When she hits the go button, the text-based site search algorithm is going to parse her search into independent variables — namely “beige,” “crewneck,” “cashmere” and “sweater” — in order to return results that might be of interest to Jane.
On the surface, this logic makes a whole lot of sense: in separating and dealing with keywords individually, a search function can cut out a lot of products that might have no place in a specific search query!
Digging a little deeper, however, means finding some glaring flaws that really hinder text-based searches.
The drawbacks of text-based searches
Perhaps the most glaring issue with text-based searches is the fact that they exclusively rely on keywords, but, more to the point, people do not. People rarely refine their thinking down to independent variables or consider every variable when searching for an ideal product. This fact alone can quickly hamper textual on-site search functions.
Using the example from above, Jane is more likely to type in “brown cashmere sweater” than a slew of keywords or variables — her brain will pinpoint the most important wants and needs and relay them as succinctly as possible. This puts the burden of narrowing things down on the search function. Moreover, it creates a gray area of possibilities that might not be considered by a text-powered search: “beige” might not be a recognized variation of brown, for example.
Along with not recognizing or properly interpreting some keywords, text-based searches also don’t have the power to distinguish tiers of importance when it comes to variables or how keywords work together. “Sweater,” “cashmere” and “beige” must be processed not only in order of importance to create viable results, but also in conjunction with one another. Putting too much emphasis on a single keyword or not linking several variables together will skew results.
Finally, because text-based searches rely exclusively on what’s typed into the search box — rather than making inferences — it’s easy to nullify key characteristics when it comes to grammar and spelling mistakes. Spelling “beige” incorrectly or failing to hyphenate “crew-neck” might mean those variables are tossed out of a search because they’re not recognized within your e-commerce site’s unique search environment or product inventory.
Natural Language Processing Bridges the Gap
Semantic search has quickly become a superior choice over keyword-based text searches because of its ability to interpret the unique needs and wants of customers. Rather than relying only on the limited information that may or may not be provided by a customer, natural language processing works to make inferences that proliferate that information.
In a semantic search, Jane’s query for a “brown cashmere sweater” will be extrapolated beyond just the words she types in:
Brown returns options classified as beige, mahogany, chocolate and any other specific types of brown that may exist for a sweater.
Cashmere brings up sweaters that might not have “cashmere” in the name, but which have that material in the blend. What’s more, 100 percent cashmere sweaters, as well as blends, will be shown.
Sweater goes beyond just long-sleeve knits, to encompass different styles and types that may be inclusive to what a sweater is.
Together, all of the above variables will be processed in tandem, instantly, returning a highly refined segment of results that’s exactly what Jane is looking for. Natural language processing means interpreting what she wants and giving her options that are as close as possible.
Keywords hit roadblocks; semantics overcome them
It’s not always easy for people to say what they want, because it’s not often we always know exactly what appeals to us! Semantic search makes it easy to narrow our options and exposes us to a close-knit group of products that exemplify our core specifications, giving us the option to make a few final refinements that lead us to the perfect product.
