Skip to main content
RichRelevance

15.03 Release Notes (Mar 30, 2015)

Discover

Separate the ​personalized and global sort algorithms for search results with new strategies.

Edit section

With two strategies instead of one, we can compare the two Search and Browse algorithms (personalized vs. global) and better control when each is used. 

Until now, sorting search results was controlled by a single strategy that returned both personalized and global (non-personalized) products. We've split the strategy into a personalized and non-personalized version, which gives us the reporting separation that clients have been asking for natively through our strategy reporting. Additionally, it gives us more flexibility: you will be able choose on a per-context basis if you want to run one or both versions of the strategy. You can also let our King-of-the-Hill engine decide which is better if you want to. Along with this change, we updated the sorting algorithm to provide more predictable results on the new strategies.

This release brings the changes to search-result sort. The same change was made to category sort in the 15.01 release

Recommend 

Updated King of the Hill algorithm more evenly distributes placement views between the best-rated algorithms at that moment.

For sites that are optimizing for CTR and attributable CTR (most sites), Thompson CTR will allow King of the Hill to more evenly distribute placement views between the best-rated algorithms at that moment. Where before one "best" algorithm would stand out with 90+% of views often, now that space will be mathematically and intelligently shared more evenly between the best competing algorithms. This is great new technology from the Lift team maximizing reward (click through) with respect to exploration (variety of algorithms). More information available here: http://en.wikipedia.org/wiki/Thompson_sampling

Similarly, the new technology for "Per-Placement KOTH" gives KOTH the extra information of which placement is being requested, so two placements are treated distinctly for better results.

Omnichannel top sellers and purchase CP strategies are now available.

When a retailer provides their in-store transactions, these will now be incorporated into the creation of omnichannel top sellers and omnichannel purchase CP strategies. Also available are offline top sellers and offline purchase CP ​strategies. 

Hierarchical Segments - (SOW) Recommend can now process multiple segments for a single user and work through the hierarchy of segments to provide segment-based recommendations at the available level of hierarchy.

Brand-based strategies bug fix

Brand-based strategies were not returning the Brand (Placeholder) inside the layouts. This has been resolved. When using brand-based strategies an email placement, the brand name now appears in the layouts through attribute placeholder ‘Brand’.

Advanced Merchandising 

Use a price range to define recommendations.

Merchandisers can now define recommendations based on a price range through the use of AND and OR operators. 

Recommendation preview expands to 50 items.  

A single Advanced Merchandising rule can support up to 100 recommended items per seed. For the recommendation clauses – merchandisers can now set up to a cumulative 100 items for recommendations.  A single recommendation clause will display previews of up to 50 items.

Check compatibility with new CNET DataSource markets.  

As of this release, all of our international instances of CNET DataSource are available for use with Advanced Merchandising.  This adds France, Canada, Nordic, and Baltic to the existing US and UK instances.

Bug Fixes

Sorting by Price

Merchandisers can now sort recommendations based on price for advanced merchandising recommendations and view the ascending/descending sort in rules preview.

AND/OR Statements

The combination of using AND/OR statement across attributes was not being interpreted correctly between the front end and back end creating unexpected recommendations.  This has been resolved.

Omnichannel

Offline transactions are now integrated with online transactions.

We have brought the in-store transaction feed in line together with our online purchase data.  As part of this refactor exercise – we are now putting all purchase information (online or offline) together in avro for model creation, the user purchases are being loaded into the UPS, and the offline transactions are being loaded into HIVE for BYOS and Analytics purposes. 

Dashboard

Site analytics now uses the category picker/browser to select categories.

The old-style autocomplete category lookup has been replaced by the category picker/browser that's improved other tools.

Build your layouts faster and more reliably with these improvements. 

  • Layout changes are are now logged in the activity log. 

  • Copied layouts now copy the placeholders along with the layout. You no longer need to recreate placeholders (layout variables) when you copy a layout.

  • Layouts now support the $PLACEMENT_NAME$ placeholder at the header/wrapper level.

Explore even more of your catalog with these catalog browser improvements.

  • Regions and channels now have a “refresh” option so you can reload the page using the new data.

  • We gave the search box more room.

  • User search wasn’t functioning properly. We fixed that.

  • The category mouse over is now cleaner and more accurate.

  • The region selection is now pre-populated/auto complete with regions for that client.

  • Fixed a bug where long category names caused some overlapping.

Client-facing site configurations and site JavaScript pages are now live for clients that have access to placements.

Both site configurations and site JavaScript can have a huge impact on your site. Be extremely careful when changing either!

 

  • Was this article helpful?