Train data import
The behavioral recommenders are trained on user behavior. Luigi's Box analytics collects co-purchases (items bought together within same session, by same user, etc.). It however takes some time from the beginning of data collection until we have enough knowledge to learn good quality recommendations. To speed the learning period, we can import a log of historical transactions from a file.
The import file must be in the json or csv format. It has two mandatory attributes, session_id
and identity
, which are used as a basis for global (anonymous) co-purchases learning. The file can contain two additional attributes, user_id
and created_at
, which, if present, allow the transaction metadata to be stored in a user profile and improve personalization.
Attribute | Description |
---|---|
session_id required
|
Any value enabling to identify products (rows) purchased in the same session. |
identity required
|
Resource identifier of the purchased product. |
user_id optional
|
Id of the user who purchased the product. |
created_at optional
|
Timestamp of a purchase used to sort purchases in time. |
Example of an import file in the json format:
{"session_id": "1","identity": "/p/123","user_id": "4", "created_at": "2023-04-22 15:04:30.12312"}
{"session_id": "1","identity": "/p/234","user_id": "4", "created_at": "2023-04-22 15:01:33.12345"}
{"session_id": "2","identity": "/p/123","user_id": "3", "created_at": "2023-04-21 00:04:38.12121"}
{"session_id": "2","identity": "/p/345","user_id": "3"}
Example of an import file in the csv format. File should not contain the header, rows contain fields in the following order - session_id
, identity
(optionally followed by user_id
, created_at
):
1,/p/123,4,"2023-04-22 15:04:30.12312"
1,/p/234,4,"2023-04-22 15:01:33.12345"
2,/p/123,3,"2023-04-21 00:04:38.12121"
2,/p/345,3