Recommendation will give customers a richer experience by a user friendly interface, from which he can directly edit and apply the dynamic settings, instantly see his changes to affect the result list.
The recommendation comes with some terms, list below are definitions :
Name - used for identification within Makaira and overview
Identifier - used for integration in the store to be able to create different recommendation blocks on different pages
Recommendation Base - defines what type of recommendation is used
- Picture Similarity - The images we have stored under picture_url_main (see Data inspector) are used to determine the most similar product image.
- Text Similarity - Based on the item description and text fields, the most similar products are determined.
- Bought together1 - Products that are usually bought together with this product.
- Behaviour history (Cart based)1 - Other people who had this in their shopping cart also bought these products.
- Behaviour history (User based)1 - Other users who bought this also bought these products.
- None - this is a request specially adapted to your question. Contact us directly if you have any questions.
- Populate with Text Similarity - This option adds more products based on Recommendation Base "Text Similarity“ if not enough results can be retrieved. Please note that this option only available if Recommendation Base is not None or Text Similarity
- Filter - With filters, you can filter down the recommendation products. You can either use concrete values or values based on the product that the recommendation is based on (input product).
- Boosting - Here you can boost products with specific values to the front of your recommendation list. It it possible to define specific rules for a field value or to use the raw value compared to the values of the other products (use value).
- Use Machine Learning - Use machine learning to effect the recommendation result. Please note that this option has effect only if you use the Makaira Tracking, enough data is collected and machine learning is booked.
- Use Ranking-Mix - Use ranking mix to effect to the recommendation result(see also: ranking-mix)
- Randomize - Ensures that the same item is not always displayed at the top of the list. Please note that only a small re-scoring will take place so that most likely no irrelevant products are shown on the beginning of the list.
The working flow of recommendation can be described in following steps :
- Create new recommendation
- Edit recommendation with these settings :
- activate / deactivate
- recommendation base
- populate with Text Similarity
- apply filter
- apply boosting
- apply extra features
1. Create new recommendation
First step is to create new recommendation, login to Makaira back end, click on left menu Recommendations -> Create
In the create new form, input name and identifier (identifier must be unique)
2. Edit recommendation, apply the settings
To apply the settings for a recommendation, you must edit it first
In the edit form, you can apply the settings, take a live preview to see how it will look like when display in store front.
The overall edit form will be look like
if inactive no recommendation products should be returned
2.1 Recommendation Base
The meaning of each recommendation is explained here
2.3 Populate with Text Similarity
With filters you can filter down the recommendation products. You can either use concrete values or values based on the
product that the recommendation is based on (input product)
2.4.1 Output Product Field
All the fields of product which can be used to filter the conditions
|equals||filter exactly the same value|
|like||filter where field contains the value|
|not like||filter where field not contain the value|
|in list||filter where field's value are in the list of selection|
|not in list||filter where field's value are not in the list of selection|
|empty||filter where field's value is empty|
|not empty||filter where field's value is not empty|
|greater||filter where field (data type number) greater than|
|greater or equal||filter where field (data type number) greater than or equal|
|less||filter where field (data type number) less than|
|less or equal||filter where field (data type number) less than or equal|
|between||filter where field (data type date) between a range of period|
2.4.3 Compare With
|Static Value||you can put in here which value should be used|
|Input product||the value will be taken from the product for which you request recommendations from makaira . this changes dynamically with the product|
Example filter: get only products with price equals 123 (Euro)
Example filter: get only products with price equals the price of input product. This allows you to create more genral recommendations, which will fit in more use cases. In this example a shop user will only see other products that have exactly the same price as his current product.
Here you can boost (move) products with specific values to the front (higher order) in your recommendation list
2.5.1 Output Product Field
The purpose of this is the same with Output Product Field, which described in 2.4.1
|use value||Special case for boosting: only for int/float fields - this boosts the product based on the value in the field (e.g. higher rated products to the front)|
|equals||product with exactly the same value will be moved to higher order|
|like||product with field contains the value will be moved to higher order|
|not like||product with field not contain the value will be moved to higher order|
|in list||product with field's value are in the list of selection will be moved to higher order|
|not in list||product with field's value are not in the list of selection will be moved to higher order|
|empty||product with field's value is empty will be moved to higher order|
|not empty||product with field's value is not empty will be moved to higher order|
2.6 Extra Features
2.6.1 Use Machine Learning
This option allows you to automatically reorder the products based on the on the specific user that visits your shop (device, location, time,..). Note that this will only apply on top of your filter and boosting configuration and only take effect on the top products. If you want to learn more about Makaira Machine Learning / Personalisation take a look at machine-learning.
2.6.2 Use Ranking-Mix
This option allows you to reuse your Ranking-Mix (ranking-mix) also on the recommendations. E.g. you don't have to reconfigure all of your general boosting.
This option adds some entropy to results. The score of each recommended product may change randomly up to 20%. As a result a user may not always see the same recommendations and you can increase the probability that he may see something that interests him. We recommend to use this option only together with some additionally boosting to make sure that the results are not completely random.
In the preview mode, you can search for a product to show recommendations. Change recommendation settings and see how the preview is changing accordingly.