AI Comes To Ecommerce Site Search
by Gareth Dismore , Op-Ed Contributor, May 8, 2017
Ask consumers what they think about artificial intelligence (AI) penetrating their day-to-day lives, and you’ll likely hear mixed emotions. On one hand, there’s something “weird” about how smart AI is getting (and how quickly) — it may not be long before our smartphones accept and answer calls for us.
At the same time, there’s something wonderful about it. While many consumers don’t understand exactly what AI is or how it works, they’re willing to accept it if in fact it will make our lives easier. Whether we know it or not, AI is already infiltrating more facets of our life than ever before, including the Web sites we shop on every day.
NLP is a type of artificial intelligence (AI) enabling computer programs to understand human speech, just as it is spoken. When applied to ecommerce site search, NLP supports a meaning-based approach, allowing shoppers to search for items just as they would ask a store associate in-person. With recent research showing consumers value easy product findability above all other Web site attributes, NLP is poised to make a very valuable contribution here.
Traditional search engines like Google Site Search, SOLR and ElasticSearch rely on basic keyword matching technology. These incorporate NLP, but the form of NLP is geared toward breaking apart long sentences in order to understand what people are looking for. While it is great for searching news articles, webpages and blogs, this type of NLP is not ideal for ecommerce, where shoppers tend to use shorter search terms. Consider a shopper looking for “white flared dress.” Traditional NLP algorithms might pick up on “dress,” but miss the important qualifiers (style, which is “flared,” and color, which is “white”).
New forms of NLP focus on “product awareness” — identifying the primary product being sought and intelligently discarding (or de-prioritizing) search results that may contain the words “white” or “flared,” but may not be a dress — white flared pants, for example. NLP understands that “dress” is the primary item being searched for, and if something is not a dress, that is a deal-breaker — no matter how many matching qualifiers there may be.
New forms of NLP are also capable of delivering “descriptor” awareness, learning from previous searches which types of words are the most important qualifiers for various products. For example, for dresses it may be style, while color is secondary. So looking at “white flared dress,” the search engine will recognize “flared” as the style and prioritize a blue flared dress over a white tube dress in the rankings.
Ability to Handle Linguistic Nuances
Another common challenge is the tendency for shoppers to use similar (although not identical) search terms, based on the shopper’s unique lexicon preferences or context. For example, one shopper may search for “one-piece blue children’s pajamas,” while another might search for “one-piece blue children’s PJs,” and yet another might search for “blue children’s onesie.” NLP has now learned to identify these items as being one and the same thing, producing relevant results regardless of the exact terminology being used.
In a similar vein, NLP now supports synonym identification, even in cases where the spellings are very dissimilar — for example, “pants” and “trousers”; “underwear” and “briefs”; “onesies” and “footies.”
NLP also identifies spelling mistakes, ensuring that a search for “red jackat” will deliver the same results as “red jacket.” Nothing annoys impatient, time-pressed customers more than to get a no results page as a search result, simply because they made a fat-finger spelling mistake when entering a search term. With NLP, commerce-oriented sites and apps can ensure their conversion rates don’t suffer due to slight human speech vagaries and spelling mistakes.
What is Coming?
The capabilities described above are just the tip of the iceberg in terms of what NLP can bring to ecommerce site search, and the overall retail industry.
While conversational commerce — where chat serves as the commerce interface — is still in its infancy, virtual assistants like Amazon Alexa, Apple Siri, Microsoft Cortana and Google Assistant are getting consumers more used to the idea of talking with a machine. Conversational commerce has the potential to set a new standard for convenience, relieving shoppers from toggling back and forth across sites and apps to gather information and make purchases.
As it moves toward mainstream adoption, NLP will play a big role, delivering even more “human-like” responses adapted to high-priority user bases, like Millennials. Even in the interim, we expect to NLP to play a role on the virtual assistant side, with shoppers asking virtual assistants for advice and recommendations first, then going to their mobile devices to conduct research and execute transactions.
In addition, with larger proportions of in-store sales being driven by online searches and research, we expect to see NLP applied to in-store operations. Store associates will be able to ask NLP-powered chat technologies where a certain item may be in the store, or alternate store locations where an inventory depleted on their own shelves may still be available.
Fifteen years ago, few people envisioned a future where a rapidly growing proportion of sales would be conducted online through digital channels. Similarly, online commerce as we know it today may look very different in the future, with chatbots disrupting sites and apps and organizations constantly striving to deliver ever-slicker, faster, more convenient experiences.
AI is gradually becoming an integral part of modern life, and NLP as it is applied to product search is one prime example that has only begun to scratch the surface of what is possible.