With this Node Type, you will be able to define an item set - filtered by ready-made algorithms called Predictors. Algorithms can be defined and named on the Settings/Algorithm Library Menu. You can choose only one predictor in one Algorithm Node. Algorithms are designed to predict how preferred would be an item, in a concrete situation, where the "preferable" would be represented by ranks and weighting.
- Select Algorithm Type
- Choose Predictor Type
- Predict Similarity
- Define context
- Sorting results by Prediction Value
Select Algorithm type:
- Row: select any of the predefined Predictors
- Custom: refer to a custom Predictor
Choose predictor type
Here you will be able to recall the Predictors from a predefined pool.
Algorithm Nodes are wrappers for the Recommendation Algorithms. There are two wrapper types which are similar, but not all of the recommendation engine Algorithms can be wrapped in both of them.
Predictor First Block
The Predictor First Block is useful if a certain algorithm has p-value only for a limited number of Items. This way it speeds up the processing because the Nodes connected after a Predictor First Block have to process only a limited number of items.
In the Property Panel all compatible Algorithms are listed. If a certain Algorithm is hovered with the mouse than its description is visible next to the list of Algorithms.
The maximum number of Returned Items can be limited manually as well. This function can be useful e.g. in case of a Popularity Predictor, when only the 100 most popular items are interesting.
By default, the selected Algorithm doesn’t take the item ID received as context property into account, but if the “Predict Similar” checkbox is checked, then the algorithm tries to use this information too. In this case the “itemId Context Property” has to be specified as well.
The rest of the checkbox is for advanced use-cases and it should be left as they are by default.
Updated almost 4 years ago