How to improve search quality?

Every Store is unique: it has different products, business processes, requirements for search.
And even if Kea Labs Search understands semantics and smartly matches products, some adjustment is needed to reflect these unique requirements of Store.
One of the things which often improves search quality is configuration of searchable content and priorities.
Sounds complicated? No worries, in this article you may find a lot of examples. But if something remains unclear, you have more sophisticated business requirements, or just need an advice - simply contact our team and we’ll help.

Data fields

Generally, in every Store has at least such information about products:

  • Product name
  • Id - unique identifier
  • Price
  • Image
  • Description

We name it as Data Fields.

In some stores there are way more fields, like vendor code, brand, model, parameters, tags, SKU’s.
These are direct fields of products, but it’s very often when other things, like categories or collection contain information which needs to be included to product. For example, name of category ‘corner sofas’ describes products very well and it's definitely a good information to be attached to product.
Another example of information - is color, which Kea Labs takes from photos, so you can search for ‘Red dress’ and receive high-relevant results.
However, making every available Data field searchable is bad idea pretty often. See details later.

Priorities

It’s natural that some of Product Data Fields describe product better and shall have higher importance. But other, needs to be used as additional information, or, even be ignored by default.
Kea Labs allows to specify the priority for each Data Field, you can easily choose which field needs to be boosted, ignored, or be scanned with a normal priority. We also have the ability to tweak priority more precisely - just ask our team to help.

Fuzzy Search, Spelling correction and Autocomplete

Another important concept is for which Data Fields you’d like to have spelling correction, or autocomplete enabled. This has a very big impact on the level of noise - how many undesired results the search brings.
Imagine what may happen when autocomplete with a spelling correction is enabled for a 1000-symbols long description: search may even suggest almost every product when you type something simple and popular, like ‘dress’. If you have ‘those shoes matches evening dresses’ in description - you’ll already get some shoes in results. That’s definitely not a desired item for User.
From another hand, it might be very useful to have autocomplete enabled for vendor code. It’s pretty often used by a store staff, especially when you have thousands of products. It allows to quickly and easily find needed items.

Examples

Well, enough of boring theory, let’s switch to examples:

Product Name

Naturally, it’s one of the most important information about product. Add it with a top priority, spell check and autocomplete enabled - pretty much in every store it's the most oftenly searchable information.

Product Description

Be accurate with this field. Some of description contain only facts and necessary information about products, on other stores descriptions might include even user feedbacks, instructions and youtube videos.
Don’t raise priority for descriptions - it may bring a lot of undesired items.

And don’t enable autocomplete or spell check - it’ll bring you tons of surprising items in results.
The rule is straightforward:
If you have informative descriptions- add them with a low priority. Otherwise we recommend to ignore it.
You can also disable search in description(and other low priority fields) by default. After this user would be able to choose on ‘Search in description’ checkbox whether such data fields needs to be searched.

Brand, Vendor and Model

For most of the stores we recommend to have them searchable with a top priority, with spell check and autocomplete enabled.
Shoes, clothes, electronics, food - pretty much everywhere brands are popular in search phases:
‘Samsung phone’, ‘Levi’s sneakers’, ‘perfume Chanel’

Vendor code, Product Id

We recommend to have vendor code to be scanned with a high priority with autocomplete - it’s useful for a store staff and allows quick navigation. You may also consider having fuzzy search enabled on this field.
If you have complicated vendor codes and would like to have other notations searchable - let us know.
i.e we can have pretty much any transformations of vendor code - i.e. remove trailing zeros, search by parts etc.

On some Stores, product ids are useful for search as well.
If your users may see product id, consider adding this into search as well. For example, customer may call to your support and ask ‘do you have shoes in a brown color available?‘. You may ask them to tell a product id from the page url, or other place, and then quickly find it and check. Tiny feature, but it saves your time and shoes your proficiency.

Kea Labs Smart Context

This is the place where power of Machine Learning gets applied.

Pretty often Products don’t have enough fields describing it, or useful information is spread on the low-important content. In order to solve this, Kea Labs Search analyzes all the products and enriches it with additional information, such as:

  • Product type - evening dress, corner sofas, laptop for gamers
  • Adjectives, describing the product - cozy, bright, convenient etc.
  • The most important attributes for this type of products - i.e. screen size or material
  • Attributes helping to differentiate product within a similar products
  • Pricing segment which adjusts further search results for User

Some of the information is available on when Kea Labs Recommendations are enabled in store, as recommendations allow much deeper products analysis.

We recommend keeping this information searchable with, at least, normal priority.