Release Summary - Apr 04, 2024 (24.07)
The following key features and improvements, along with bug fixes, have been released in Algonomy CXP products in the release version 24.07 during Mar 22, 2024 - Apr 04, 2024.
Enterprise Dashboard
Allow restriction rules to be applied to Advanced Merchandising results
We have introduced a significant enhancement to Advanced Merchandising rules, allowing for more flexibility and relevance of recommendations. Optimization managers can now build an Advanced Merchandising rule universally and apply Recommendation Restriction or Recommendation Boosting rules on the results.
This can be useful if wanting to add a temporary rule to only recommend products that are in the current promotion. You no longer have to add this as a criterion to each Recommendation Group in each Advanced Merchandising rule. Instead, you can create a single Recommendation Restriction rule that will be applied on any Advanced Merchandising rule that has the new option enabled.
This also allows for applying Restriction or Boost rules on localized attribute values, ensuring recommendations are pertinent to the specific region being viewed.
Enabling the option: A new checkbox option, " Enable system filters and rules," has been introduced under the rule name in the UI, with a tooltip explaining its function. When the checkbox is enabled, it applies Recommendation Restriction, Recommendation Boosting rules, and system filters to the Advanced Merchandising rule results.
Dynamic Rule Application: Upon evaluating a placement, the initial Advanced Merchandising rule result set is processed, and then merchandising rules along with general filters are applied to refine the recommendations further.
Jira: ENG-27722
Enterprise Dashboard, Data Engineering
New Reporting Features
We are thrilled to announce the addition of new reporting features designed to deepen your insights into customer behavior and aid in refining your strategy.
Co-occurrence Report
This new tool analyzes the frequency at which product categories are purchased together, offering valuable insights into customer buying behavior. It enables you to:
- Identify complementary products by understanding which categories are frequently purchased together.
- Optimize product placement based on the attach rate, informing decisions about in-store or online product placement.
- Develop targeted promotions by leveraging insights to bundle frequently co-purchased products.
Features include the ability to choose a specific product category for analysis, view the co-purchase count within a 30-day period, and calculate the attach rate to determine the percentage of orders that include related categories.
Product Placement Report
Introduces a detailed view of product recommendations within specific placements, allowing you to evaluate the performance of products in these placements. It facilitates data exploration at both the placement and product levels, helping you make informed decisions regarding product placement strategies.
The report is accessible via a new dedicated tab within the existing Merchandising Report, ensuring easy navigation.
Scorecard Report
Aims to review catalog health periodically by identifying any missing categories, products, non-recommendable products, etc., thereby helping to enhance the catalog and recommendations. It includes checking for catalog exposure, missing categories, missing or non-recommendable products, pages without user sessions, and product exposure metrics like products not viewed/added to cart/purchased from the catalog.
Jira: ENG-27934
Other Feature Enhancements
The following feature enhancements and upgrades have been made in the release version 24.07 during Mar 22, 2024 - Apr 04, 2024.
The following feature enhancements and upgrades have been made in the release version 24.07 during Mar 22, 2024 - Apr 04, 2024.
|
Jira # |
Module/Title |
Summary |
General Availability |
|---|---|---|---|
|
|
Science: Shop the Look - download images (create an Airflow pipeline for STL) |
We have introduced an Airflow pipeline to automate image downloads for the "Shop the Look" (STL) feature, improving the model's accuracy and user experience. This pipeline is tailored to fetch images based on specific configurations for STL model seeds and recommendation visuals. Upon downloading, the STL job is executed on AWS GPU for enhanced processing, with the results stored in HDFS and bounding box coordinates saved in the database. |
04-Apr-24 |
|
|
Enterprise Dashboard: Recs Test Drive - A New Tool for Optimization Managers |
The Recs Test Drive feature is now available for optimization managers, offering a preview of the recommendations users will see on various placements. This tool allows managers to assess the system's configurations and ensure recommendations align with expectations. Key features include viewing recommendations by placement name, testing the impact of different User IDs, selecting specific strategies, and understanding which rules are triggered under certain conditions. Managers can also specify product seeds, choose channels, and set regions to see how these factors affect the recommendations. Note: This feature is currently in beta. |
|
|
|
Recommend: Ability to create response URLs on algorecs.com domain facilitating the use of new domain from runtime. |
We have updated our system to dynamically adjust response URLs, including click-tracking links, based on the originating request domain. Previously, responses defaulted to recs.richrelevance.com, even for requests from algorecs.com, due to the vanity URL setup or fallback configurations. Now, without a specific vanity URL configuration, the rrserver generates response URLs that match the request's domain. This means requests from algorecs.com will have response URLs on the same domain. |
04-Apr-24 |
|
|
Find, Streaming Catalog: General Support for LLM |
We have introduced general support for Large Language Models (LLM), providing tools to define prompt templates, REST services for fetching LLM results based on template attributes, and a Python library dedicated to these operations. This enhancement ensures that all APIs related to LLM interactions are performing optimally, marking a successful integration of LLM support into our system. |
04-Apr-24 |
|
|
UPS: Develop a notification system for missing online UPS events |
We have implemented a notification system aimed at identifying and alerting on missing online User Profile Service (UPS) events across all sites. This system meticulously monitors for anomalies, ensuring any synchronization issues are swiftly detected and reported directly to Datadog. |
04-Apr-24 |
|
|
UPS: Enhancements to recommendable Product Count Alert |
We have made significant enhancements to the Recommendable Product Count Alert system. Key improvements include transitioning the Datadog metric from a Gauge to an Event Type for more dynamic monitoring, enriching alert messages with detailed information, and adjusting the execution frequency to daily. |
04-Apr-24 |
|
|
Find: Prompt TTL and version JSON Blobs db changescripts |
We have implemented database enhancements to include Time-To-Live (TTL) for prompts and introduced JSON blobs for version management. These JSON blobs are specifically designed to hold Large Language Model (LLM) provider configurations, such as temperature settings for ChatGPT, optimizing result fetching based on provider-specific parameters. |
04-Apr-24 |
|
|
UPS: Add a per site Config to enable Glass views |
We have introduced a per-site configuration option to enable Glass views, ensuring that the User Profile Service (UPS) stores data only when this feature is activated. |
04-Apr-24 |
Bug and Support Fixes
The following issues have been fixed in the release version 24.07 during Mar 22, 2024 - Apr 04, 2024.
|
Jira# |
Title |
Summary |
General Availability |
|---|---|---|---|
|
|
Enterprise Dashboard: Fix Display of Renamed Configurable Strategies in Merchandising Rules |
We have addressed an issue where the original names of renamed configurable strategies persisted in associated merchandising rules. Now, when a configurable strategy is renamed, all related merchandising rules—Do Not Recommend, Only Recommend, and Boost—accurately reflect the strategy's current name. |
04-Apr-24 |
|
|
Recommend: Configurable Strategies Diversify by Category |
We addressed an issue in Configurable Strategies involving the Attribute Top Sellers or Attribute Top Views model. When utilized with a fixed attribute seed and paired with category or brand diversity filters, the first product was incorrectly filtered out. This has now been corrected. |
04-Apr-24 |
|
|
Recommend: Price filters not working in recsForPlacementsContext API for standard strategies |
We fixed an issue where the price filters were not working as expected in the recsForPlacementsContext API. |
04-Apr-24 |
|
|
Data Engineering: Category CoPurchase Metrics - Attach Rate is Incorrect |
We resolved an issue with the Category CoPurchase Metrics where, following an enhancement to consider all categories instead of just primary ones, the attach rate was reported inaccurately across all co-purchases within a category. This has now been corrected. |
04-Apr-24 |
|
|
Streaming, Recommend: Some itemstore instances have missing data caused by possible dirty read at MapDB3 |
We have addressed an issue in the streaming itemstore, crucial for streaming recommendations, where product attributes were inaccurately missing due to potential dirty reads in MapDB3. Implementing forced commits after batch processing has ensured consistent data integrity, effectively rectifying the problem. |
04-Apr-24 |