Hello Community,
I'm testing different cases of using ML of type "Recommendation prediction" based on documentation https://academy.creatio.com/documents/administration/7-16/recommendatio….
Described solution is working on purchases of similar products data set. How to set model to learn prediction based on similar customers purchases ?
Is there any description or example how to use advanced tool (Query metadata for selecting additional training data, Query for selecting additional training data) on this type of ML model (Recommendation) ?
Regards,
Marcin
Like
Marcin Kott,
Advanced parameters are merged with parameters that are formed by standard tools and as a result are formed one common set of parameters. Moreover, in the advanced settings, you can also specify additional filtering parameters.
For recommendation models, no "additional" parameters need to be passed, since the models are trained on transaction data, or any other interaction of the client with the product, for example, the product in the order. The model will not accept any other parameters.
Extended selection is needed for these models, for example, in order to combine data on purchases of goods and views of goods on the site. Then a selection is written that somehow aggregates this data, but as a result, it is necessary to still get a selection of the type - client, product, number.
In recommendation case in "inputs" tag i may add Id and eg. Account Country, Account Category ? - It is impossible to divide this way, since the selection should contain only Id (just an identifier)
user_id (a user)
item_id (something purchased by the user)
value
Finally, there are actually two options for recommendations – selection by standard tools or completely form it using a custom query.
Best regards,
Norton
Hi Marcin,
To set the model to learn prediction based on similar customers purchases you should add the following columns to the query:
Id (just an identifier)
user_id (a user)
item_id (something purchased by the user)
value (how much did the user purchase, but you can leave it 1 if it's important to see the fact of a purchase)
You should add those values to additional metadata as it is shown here:
https://academy.creatio.com/documents/technic-sdk/7-16/how-implement-custom-prediction-model
You can find more information regarding queries here:
https://academy.creatio.com/documents/technic-sdk/7-16/creating-data-queries-machine-learning-model?document=administration
Unfortunately, we don't have a separate article regarding this type of ML model (Recommendation).
Thank you.
Bohdan Zdor,
Thank You for explanation. To better understand this tool please clarify me some concerns:
Is this “advanced tools to add columns” overrides basic training data set resulting from “Recommendation dependency” query or this is additional data which are corellared with basic query ?
I mean .. “Query for selecting additional training data”, “Query metadata for selecting additional training data”.
I yes, then what is principle to building additional select ("new Select") to maintain relation between 2 queries (basic and advanced) ? Eg. that selected object should be the same (“Recommendation dependency” = "select .. From x"), what else ?
“Query metadata for selecting additional training data” – In posted documentation
additional columns are set:
{
"inputs": [
{
Id
},
{
attr1
},
{
attr2
}
],
"output": {
attr3
}
}
In recommendation case in "inputs" tag i may add Id and eg. Account Country, Account Category ??
What I have to add then in "output" tag ?
In Your example I understood that as a input I have to add
Id -> Object Id
user_id -> Account Id
item_id -> Product Id
value -> Quantity
when eg. Query data set is a “Product in Quote”, right ?
Regards,
Marcin
Marcin Kott,
Advanced parameters are merged with parameters that are formed by standard tools and as a result are formed one common set of parameters. Moreover, in the advanced settings, you can also specify additional filtering parameters.
For recommendation models, no "additional" parameters need to be passed, since the models are trained on transaction data, or any other interaction of the client with the product, for example, the product in the order. The model will not accept any other parameters.
Extended selection is needed for these models, for example, in order to combine data on purchases of goods and views of goods on the site. Then a selection is written that somehow aggregates this data, but as a result, it is necessary to still get a selection of the type - client, product, number.
In recommendation case in "inputs" tag i may add Id and eg. Account Country, Account Category ? - It is impossible to divide this way, since the selection should contain only Id (just an identifier)
user_id (a user)
item_id (something purchased by the user)
value
Finally, there are actually two options for recommendations – selection by standard tools or completely form it using a custom query.
Best regards,
Norton