Skip to main content

Relevant Facets for search queries

· 2 min read
  • vorselektierte Filterwerte
    • per KI vorberechnen
    • über Admin Tool pflegbar machen

Precompute filters for each category​

  • for each category facets are precomputed based on statistical data
    • the attributes with the most values for a categories are good facets
    • the usage of the facets are tracked, facets that are used more often are more important
    • TODO: position bias - facets on top positions are clicked more often because of the position
    • TODO: facets that are never shown to the user have no chance to get clicks
  • a classifier is trained,
    • input: query (use tracking date)
    • output: category
  • for each query, the classifier is called and based on the classified category the facets are used

Precompute filters based on historical queries - query to query vector search​

  • for the most historical queries, facets are precomputed
  • it is not possible to precompute facets for all potential queries a user can use
  • the idea is to use a vector distance, to find the nearest query and use the facets of this query

Automatically compute filter on query time​

  • index the filter for each document - field: filters
  • for each query, facet on the field filters to determine relevant filters
  • show the facets for the search result

Advantages​

  • the filters can be computed automatically without manual effort

Disadvantages​

  • 2 sequential search queries are required, which doubles the search time
  • the relevant filters are based of statistical values and might be not optimal for every case

Automatically select filters for specific queries​

  • manually define selected filter for a query
  • try to find filters based on statistical data

Disadvanatages​

  • similar queries

Facet Types​

  • Slider
  • Slider with histogram
  • Color picker
  • date picker
  • shoe size picker

Ideas​

  • show on the facet value the image as a preview of the first product when the filter is selected