Correlated and Similar Item
Correlated items and Similar items is a part of the standard AGR system, and it utilizes machine learning techniques to identify such products.
Correlated items algorithm uses sale transactions or point of sale data to identify positive correlation/halo effect between different products. A typical example of this is hot dogs and hot dog bread. If the sale of hot dogs increases, for example, because of a price cut or because of the weather, then the sale of hot dog bread will increase by almost the same percentage.
To find Similar items the algorithm uses various descriptive data about each item, e.g. name, description, manufacturer, ingredients, product group. The similar item algorithm compares all items based on the data provided and finds their similarity. An example of similar items is different types of ketchup.
The Correlated items and Similar items workspace is found under the ‘More’ tab -> ‘Correlated and Similar items’ .
1.1 Correlated item
This view displays all correlated items. The drill-down groups correlated items by item group level 1. The columns in this view are as follows:
- Parent item No & Name - Item number and item name for the item that has a correlated items. This item is referenced as the “Parent Item”.
- Correlated item No & Name - Item number and item name for the correlated item.
- No. of Locations - Number of locations where this item set of {parent item, correlated item} is distinguishable. Purchase behavior can differ between store locations and therefore some correlated items are not detected at all locations.
- Confidence - This is the measure of the likeliness of buying item B when already having item A in the cart. E.g. when having confidence = 0.29, it means that in 29% of cases when a customer buys item A, he also buys item B.
- Lift - This measure assists us in evaluating how relevant a particular correlation is between two items. The definition of lift says that if the lift value is higher than 1, it means that the occurrences of two items are dependent on one another. The higher the lift, the more relevant the correlated items are.
- User Name - The name of the user that created this correlated item set. When ‘System’, this item set was calculated by the correlated algorithm.
- Delete or Ignore - Action buttons ignore or delete. When created by the ‘System’, the correlated item set can be ignored. When created by a user, the correlated item set can be deleted.
Correlated Item Example
Lets look at one example and inspect the correlated items in the following picture market inside the red box. Here the parent item is NOOMA Organic Pre Workout Drink - Dragon Fruit and it is correlated to two other NOOMA drinks, Lemonade and Pinapple Mango. For this particular company the NOOMA drink is a very popular product and is sold frequently (like e.g. milk or bread in grocery stores). Lets look at what confidence and lift tells us.
- NOOMA Dragon fruit –> NOOMA Lemonde: When having NOOMA Dragon Fruit in the cart, in 85.41% cases the customer also buys NOOMA Dragon Fruit. The lift value of 2.5928 (which is higher than 1) tells us that this correlation is relevant.
- NOOMA Dragon fruit –> NOOMA Pinapple Mango: Very similar correlation as above, when having NOOMA Dragon Fruit in the cart, in 80.98% cases the customer also buys NOOMA Pinapple Mango. The lift value is 2.0365, which tells us that this correlation is relevant.
- NOOMA Dragon fruit –> Gatorade Super Shake Vanilla: When having NOOMA Dragon Fruit in the cart, in 24.04% cases the the customer also buys Gatorade Super Shake. So, this item set is not as frequently bought together, but however the lift is very high, with value 18.7431. This means that when NOOMA Dragon fruit is in the cart, it is very relevant and likely that Gatorade shake will also be bought.
Create new correlated item
Users can themselves add a new correlated item set.
Press the plus icon at the top right and insert the parent item no/name and then the correlated item no/name.
You only need to fill out those two columns, the other columns will be automatically populated.
Press the green checkmark ‘Confirm’ to confirm the new item set.
Column ‘No. of Locations’ gets the number of locations that are in common between the two items, columns ‘Confidence’ and ‘Lift’ get value -1, and ‘User Name’ is populated with the name of the person that created the item set.
Delete and Ignore functionality
There are two action buttons - ‘Ignore’ and ‘Delete’.
Ignore button is used when a particular correlated item set found by the correlated algorithm (where user name is ‘System’) is not wanted. When ignoring an item set, it is removed from view 1.1 and added to view 1.2 Ignored Correlated item.
Delete button is available when correlated item set was created by a user. When clicking delete, the item set will be deleted. If you accidentally deleted an item set, you can simply create it again.
1.2 Ignored Correlated item
Here is an overview of all correlated item sets that have been ignored by users.
If you no longer need a certain correlated item set to be ignored, press the blue arrow in the right-most column.
This will add the correlated item set back to the Correlated item list, so they will appear in view 1.1 Correlated item.
2.1 Similar item
This view displays all similar items. The drill-down groups similar items by item group level 1. The columns in this view are as follows:
- Parent item No & Name - Item number and item name for the item that has similar items. This item is referenced as the “Parent Item”.
- Similar item No & Name - Item number and item name for the similar item.
- No. of Locations - Number of locations where this item set of {parent item, similar item} is distinguishable. Product assortment can differ between stores, therefore some similar items are not present at all locations.
- Score - The similarity score can take values from 0 to 1, where 1 means the two items are the same when looking at the data that was fed into the algorithm.
- User Name - The name of the user that created this similar item set. When ‘System’, this item set was calculated by the similarity algorithm.
- Delete or Ignore - Action buttons ignore or delete. When created by the ‘System’, a similar item can be ignored. When created by a user, a similar item can be deleted.
Create new similar item
Users can add a new similar item set, which can be useful if e.g. the algorithm does not pick up a certain similarity that is common knowledge to humans.
Press the plus icon at the top right and insert the parent item no/name and then the similar item no/name.
You only need to fill out those two columns, the other columns will be automatically populated.
Press the green checkmark ‘Confirm’ to confirm the new item set.
Column ‘No. of Locations’ gets the number of locations that are in common between the two items, column ‘Score’ gets value -1, and ‘User Name’ is populated with the name of the person that created the item set.
Delete and Ignore functionality
There are two action buttons - ‘Ignore’ and ‘Delete’.
Ignore button is used when particular similar item found by the similarity algorithm (where user name is ‘System’) is not wanted. When ignoring a similar item, it is removed from view 2.1 and added to view 2.2 Ignored Similar item.
Delete button is available when similar item was created by a user. When clicking delete, the item will be deleted. If you accidentally deleted a similar item, you can simply create it again.
2.2 Ignored Similar item
Here is an overview of all similar items that have been ignored by users.
If you no longer want a certain similar item set to be ignored, press the blue arrow in the right-most column.
This will add the similar item back to the Similar item list, appearing in view 2.1 Similar item.
