Jan 30, 2019
For our multi-shard merchants who have JSON faceting, the facet calls were failing in the dashboard, causing attributes to not show up in the ranking editor. A fix was made for this issue and an Out of Band release was pushed to production on Jan 28.
A minor fix was made to the UI where large chunks of text overlapped outside of the border.
An issue was reported where when a user tried to save a second variant for a category ranking rule, the "Save new variant test" modal throws an error. The variant was actually being saved, but the Rules Variant modal didn't pop up to show the user that the variant was actually saved. This has been fixed.
Our EU dashboard machines for analytics went down because our Redshift cluster was tagged as "prod" whereas the analytics dashboard only accepts the tag to be "prod_eu". Our Engineering team made the correction in the properties file and the fix has been verified in production.
With this release, we have made several updates to the Ranking Diagnostics feature.
Query Processing: we've added a new section that provides information on how the query was processed by the BloomReach engine. Features such as Autocorrect and Query Relaxation are designed to reduce the number of null result pages when your user's queries don't exactly match your product content. If a user executes a category, PID, or category ID search then query processing is not applicable and will not show any values.
- Is this query autocorrected? - This field will say "yes" if a query is being autocorrected and "no" if it isn't. If a query is being autocorrected, this field will also display the term that the original query is being autocorrected to. A query is only autocorrected when the original query returns null results.
- Is this query relaxed? - This field will say "yes," if a query is relaxed and "no," if it isn't. If a query is relaxed, this field will also display the value for what the original query was relaxed to (eg: “black” - optional, “hiking” - optional, “boots” - required). A query is relaxed when the original query returns null results and no autocorrect match is found. You can hover over the information icon in the dashboard for more details.
- Synonyms Applied - This field lists all synonyms that are being applied to the query and now displays both the left hand side and right hand side of the synonym where previously it just showed the right hand side. This should make it easier to find the source synonym that could be introducing products into the recall set.
Performance metrics in the Ranking Diagnostics Dashboard have also been simplified into the following three components:
- Performance Score: Normalized metric based on user behavior signals such as revenue, conversions, add-to-cart, views etc. generated by a specific product for a specific query/category.
- Relevance Score: Based on BloomReach Semantic Engine and Term Match to product feed data (e.g. title, category, brand etc.)
- Total Score: is the metric based on which products are ranked for a specific query/category. It is a combination of the Performance (User Behavior) and Relevance Scores. The Performance score has a much heavier weight on the Total Score than the Relevance Score.
Query to Feed Term Match:
Once you click into the product for a specific query, at the bottom of the product you will see a section called "Searchable Feed Data". This table will show you where the user query (or synonyms for the user query) are matching to content in the feed. This section can be used to answer questions like "Why is this product showing up for this query?"