- 06 Oct 2023
- 2 Minutes to read
Monitor Algorithm Health
- Updated on 06 Oct 2023
- 2 Minutes to read
An algorithm functions by taking in user events and product catalog information, processing them, and producing a curated list of recommended products. When evaluating performance, two crucial factors influencing recommendation quality are:
- The volume of user event data processed by the algorithm,
- The quality of the product catalog structure (such as diverse attributes and non-missing mandatory attributes).
To ensure accuracy and quality in recommendations, each algorithm requires a minimum threshold of user events and a minimum number of eligible products in the catalog (products with necessary attributes). Essentially, the Algorithm Accuracy is determined by examining the quantity of user events and eligible products in the catalog.
How to monitor the health of recommendation algorithms
Assessing the health of an algorithm involves examining the breakdown of "which mandatory user events are missing" and evaluating "how many eligible products are available in the catalog." This transparency in understanding missing events and catalog eligibility provides insights into the performance and robustness of the algorithm.
If there are multiple locales, any problematic locales are highlighted in a yellow ribbon attached to the top bar for quick identification. This design feature enhances visibility and allows for easy monitoring of the health of your algorithms that are working for different locales.
On the left side of the screen, you'll find the real-time status of your product catalog and user events data. If either of them registers as "Low," troubleshooting becomes imperative. To check your product catalog, you can visit Product Catalog Management Panel on InOne.
The red and green icons next to the algorithms signify the live accuracy status of the algorithm. If it shows as "Low" (in red), it indicates that the algorithm may result in fallbacks, leading to diminished accuracy in recommendations.
When you click on an algorithm from the listing table, you will see its respective details.
- Recommendation Accuracy: Reflects whether the algorithm has sufficient data to generate accurate results without fallbacks.
- Configurations: Specifies the chosen look-back period and user event sources utilized by the algorithm.
- Page Type: Identifies your pages recommended for using this algorithm, based on compatibility. Also mentions other page types where this algorithm can be applied.
- About: Offers a brief explanation of the algorithm's functioning and methodology.