Query-time Boosting

Luigi’s Box offers support for query-time boosting, allowing you to define conditions that enhance the ranking of products by assigning them higher scores.

For example, entering “fender” into two different search boxes can yield different results: one configured to prioritize results from the “Music” category, while the other prefers the “Sailors” category. The only distinction is in the prefer API parameters they utilize — one uses prefer[]=main_category_lvl_1:Music, the other uses prefer[]=main_category_lvl_1:Sailors.

Query-time boosting via the prefer parameter is supported across all Luigi’s Box services:

• Autocomplete API
• Autocomplete.js via the Prefer configuration option.
• Search API
• Search.js via the Prefer configuration option.

Boosting versus filtering

Boosting (or preference) should not be conflated with filtering. While boosting influences the ranking order of results, it does not impact the presence or absence of a product in the search results. On the contrary, filtering acts as a pruning mechanism for the result set. If a product fails to meet the specified filters, it is designated as “not found.”

For instance, consider a search for “fender guitar.” A search box which prioritizes “Sailing” will still display results for the guitar, albeit with a different ranking order. In contrast, a search box that filters by “Sailors” will yield no results, as it exclusively filters products from that category.

Boosting versus personalisation

Query-time boosting is similar to personalisation in a way that the basic ranking model is being adjusted on-the-fly. The main difference is that personalisation features adjust the ranking dynamically, considering the user’s learned preferences and historical interactions. Query-time boosting on the other hand relies on an external instruction where a domain expert has to configure which results should be preferred.

In most cases, personalization features effectively cater to user preferences without the need for query-time boosting. Nevertheless, query-time boosting can be valuable in these situations:

• Gender-based Boosting in Fashion Domains. Boosting results based on the user’s gender when the user is logged in and their gender information is known.

• Category-specific Boosting. Boosting results based on the currently browsed category. If a user initiates searches within the “Sailors” category, the system can be configured to explicitly boost results from that category.

It’s essential to note that using query-time boosting and personalization concurrently is feasible, as they are not mutually exclusive.