AI-Synonyms

AI-Synonyms

  • updated 11 mths ago

By default, the search functionality that is provided through Hello Retail, will search for matches within your product data, such as in the title, brand, description, keywords etc.  and will then rank those results, based on how much the query resemble the data. In some cases though, users will search for something that does not have any close resemblance to say, what the products are called, or to the naming of the categories, which the products have been placed inside etc.

So, as a way to minimise the amount of searches that does not find anything, clients can either manually add specific synonyms for each individual query, or have our AI-based tool handle this automatically, by enabling the feature called "Use AI-Synonyms", as shown in the screenshot below.

Note: Some search terms with no results may still not be solved with AI-Synonyms. These could for instance be product serial numbers or queries with no detected alternatives. 
It should also be noted, that we do not consider AI-Synonyms to be a replacement for - but rather a supplement to - manually adding synonyms, to improve the performance of your  search configuration at Hello Retail. Often you will have domain- and customer specific knowledge, that cannot easily be handled by automation. This article concludes, by pointing to how you can get insights into the performance of this tool, in order to have a basis for further, manual adjustments.

Enabling the feature

The feature can be found under Search > Search Engines > Edit search settings

Once this has been enabled, Hello Retail will begin to look at the individual searches that are made on your website, that does not return products. This is something that is continuously running in a queue-system in the background - so you should not expect to  see improvements for newly made zero-results searches right away, as this process will take some time.

Through the use of AI, Hello Retail then establishes a series of alternative search terms to the query, that did not find anything previously. These alternative search terms are then tested up against the product data in your database, upon which the products that are found to be a match, gets the initial search term added to their data.  Put differently, the products that where found in the search, using one of the AI-generated alternative terms for the zero-result query, gets the zero-result query added to its own product data over time - so the next person search for this, will find the product as well. 

Hello Retail does not provide any way of directly influencing how the AI is looking at words and sentences, what data-fields it is including, how these are weighed etc., so the feature can not be configured. If "Use AI-Synonyms" have been set to "Yes", but is not found to be producing the desired results over time, it can simply be disabled again (by toggling the feature to "No"), after which the alternative search terms, that have already been added to the information in the Hello Retail product database, will be disregarded in the search.

Styling the results

Reach out to technical support of Hello Retail, if you require that the output provided by this feature should be styled in a certain way, as to distinguish it from the normal search output. This could for instance be by using a specific headline for AI-based results, or by adding a  badge or similar to the products found though this part of the search logic. As a way to make it more obvious to your customers, why one or more products might have been found by the search engine, even though their data might not appear to relate to the search words used, we suggest that the search template (wherein the rendering and styling of the  search results happen) is updated with this in mind, when enabling this feature.

Performance

While the proces of adding manual synonyms, typically begins by inspecting the search statistics, observing the most commonly typed queries with no results, and then creating your synonyms from that, enabling the AI-Synonyms will make this process a bit different. As the essential objective of this feature, is to avoid having zero-result searches altogether, after enabling it, the view below should over time get less and less populated:

So, in order to get a feel for how the automation is helping your visitors find items, you should navigate to the section below, called Total Searches:

In this view, you can then:
1) Filter the results on 'AI-synonym product'.
2) Sort the searches made, on 'Clicks', in a descending order.

This will provide you with a list of queries that, at the time when being searched for by one of your customers for the first time, have had no results, so an AI-Synonym have been inserted. The statistics then depicts how the subsequent visitors have benefited (or not) from typing this same query. With this information you would now be able to add a manual synonym, to support the AI-Synonyms, that might not be performing satisfactory for the users.

Like Follow
  • 1 yr agoLast active
  • 227Views
  • 1 Following