Keyword search is very good. It is used everywhere because it is very powerful, able to search through data and find items of interest. Search engines such as Google have used the data from billions of searches to train sophisticated statistical algorithms that continue to improve their search. And they have “trained” their users to try a search, look at the results and, if the results are not those desired, change the keywords and search again.
When searching the entire web, we assume that the relevant information is out there somewhere, if we can just figure out the right search words. So, we try several times before giving up.
When shoppers are in your e-store, though, they are not searching the entire web, they are navigating just your store. And they rightfully expect your navigation, search and discovery tools to know everything about your products and help them to find the perfect item to purchase, just as they expect that level of intelligence and competence from your sales associates in your brick and mortar stores or your customer support people in your call center.
What intelligence in the e-store would help the shopper find and purchase items?
Perhaps the answer is clearest through examples of what shoppers might ask for in various stores.
- Apparel store: “a warm coat made of natural materials”
- Shoe store: “dress heels in a dark color”
- Office supply: “fine point pencils”
- Electronics: “a computer good for travel”
- Department store: “gold shoes” or “gold earrings”
The keyword search, however, is only as good as the keywords match to the product database. Apparel materials may be listed as cotton, wool, suede, rayon, polyester, faux leather, etc. Shoes may be available in black, brown, cordovan, blue, white, red, pink, gold, burgundy, etc. And some pencils may be labeled as fine point, but others as 0.5mm. What can a keyword search do?
The mismatch between the user and the product database
These shoppers are all describing what they are looking for in very clear and reasonable language and a sales associate in any store would know exactly how to help them find what they are looking for. The problem for the online search is that the shoppers’ language does not always match the specific words that the manufacturer or retailer has used to describe the products in the catalog. The sales associate is using a combination of human knowledge and domain-specific knowledge to interpret the shopper requirement into the specifications and features of the store’s products:
Examples of human knowledge:
- What materials are natural?
- What colors are dark colors?
- For a shoe, gold is (almost always) a color, but, for an earring, gold is a material.
- 0.5mm pencils are also known as fine point.
- A computer for traveling would be a small, light laptop or a tablet.
How do we put intelligence into the e-store?
The above examples show that shopper queries need to be interpreted using a combination of human knowledge and domain-specific knowledge. By integrating that knowledge with our semantic understanding language technology, we can understand the shopper and interpret the shopper’s queries appropriately to find matching items in the store. In essence, we are using advanced algorithms to emulate a good salesperson. The result is a better experience discovering products in the store for the shopper and higher conversion rates and sales for the retailer.