Customizing Retail Subdomain Data

Retail Domain Data

For each subdomain in the retail domain users can upload the following data files:

  • SKU data

  • Filters data

  • Synonyms data

  • Language data

Additionally in ‘custom/miscellaneous’ the following data can be added

  • Navigation targets data

SKU data:

CSV Structure:

  • Search will function on the SKU Name, Product Type and Brand

  • If the SKU ID is provided then upon search all SKU IDs which contains the SKU name, product type and brand will be provided

  • If there are certain tokens (part of words) such as ‘Cadbury’ in SKU name ‘Cadbury Oreo cookies’ and the brand provided is also ‘Cadbury,’ then ‘Cadbury’ will be returned as brand only.

Synonyms data:

CSV Structure:

  • Slang’s backend will use the synonyms to recognize different variations of the same key. During inference if an end user speaks the synonym, Slang’s client api response data structure will return the key. For example if a synonym of ‘potato’ is ‘aloo’, then if the end user speaks ‘aloo’, it will be recognized and the return value will be ‘potato’

  • The same file can be used to add synonyms for elements in the filter data or the SKU data file. The ‘type’ is to be used here can be one of:

    • sku_name

    • product_type

    • brand

    • category

    • sub_category

    • filter name

    • filter_value

Filters Data:

CSV Structure:

Filter data corresponds to the type of filters that can be applied on a list of products in an ecommerce app. Examples could be a filter to apply on food products: ‘Allergen info.’ The filter name would be ‘allergen info’ and the values could be ‘lactose free,’ or ‘contains milk’.

If a filter value is the same as a token available in brands or sku name in the sku data, the filter value would be given preference.

The filter name will also be used in sentences such as, ‘Remove allergen info filter.’

Language Data:

CSV Structure:

Language data provided here would be used to augment the existing translation and speech-to-text models of CONVA.

The key is the English language version of the word while the Translation column is for the corresponding word in the alternate language. The language column is to specify which language the translation is in, ‘Type’ column is to specify which type this row belongs to. The types list is the same as that mentioned for the ‘Synonyms Data’ above.

To help augment the speech recognition of CONVA in English language, add the English word in the key column, language ‘en-IN’ to the language column and leave the translation column blank.

CSV Structure:

Navigation data provided here would be used to augment the list of possible targets for in-app navigation. Suppose, a navigation target within an app is the ‘cart.’ By adding ‘cart’ to the navigation targets, sentences such as ‘Take me to cart’, will be supported.

In addition to the targets, a developer may provide corresponding URLs for the targets. The assistant will return these URLs in the response.

Last updated