In order to be able to filter by your own metadata, you need to make some selections during the model preparation stage. You will learn everything about this in this article.
You start by setting up your model as always with our wizard. Your uploaded file may look like this:
In this example we have reviews of airline trips. The customer review is our text column, and the review date is our date column. All additional columns may now be imported as metadata columns. There are two types of metadata:
-
numerical metadata is any column that contains numbers that have an intrinsic ordering to them. Examples would be any type of ratings, the age of a user, the price of a product and so on. Numbers, such as telephone numbers or user-ids, that do not have an intrinsic ordering to them do not represent numerical metadata.
-
categorical metadata is any column that defines categories across your data. In our example this would be the airline that was used, and the type of traveler. Other common examples would be locations such as countries or cities or job position. As a rule of thumb, if your data are not numbers with an intrinsic order, it should be interpreted as categorical metadata.
In our example, we would like to select the airline and traveler type columns as categorical metadata and the overall-rating column as a numerical column. We can do this in the next step of the wizard.
We have already selected the customer_review and the review_date column as the content and date column, respectively.
When selecting the metadata columns, you will be shown a list of common types of metadata, such as age, brand, city and more. If your type appears, you can select it. Otherwise you always have the option to select "Other - Numerical" or "Other - Categorical".
Here we show our whole selection:
Note that any column not selected in this step, will be dropped for further analysis.
We then validate your selection and the model training can begin.
We hope this article helps you to get the maximum out of your metadata.
Thanks for reading!