Cognitive Approach For Personalised Recommendations

Client Background

Client is an Indian online fashion retailer that sells apparel, footwear and accessories only for the 18-35 age cohort.

Business Need

Create algorithm(s) to arrange products on listing page for various categories/subcategories in likelihood of getting purchased/add to cart/clicked.

Factors driving this business need:

  1. Client doesn’t use any personalization on its listing pages (category/subcategory) due to which visitor has to scroll down to discover relevant items which isn’t great for user experience
  2. Product hierarchy is static and same for every visitor irrespective of previous visits/purchase/preferences and other recent actions

There are three top level categories with subcategories on client’s website:

  1. Clothing: T-Shirts, Formal Shirts, Casual Shirts, Casual Bottoms, Formal Trousers etc.
  2. Footwear: Formal Shoes, Casual Shoes, Sports Shoes
  3. Accessories: Bags & Wallets, Belts
Solution

Since each category/sub-category had distinct features, we developed recommendation algorithms at visitor level specific for each category/subcategory. Creation of these algorithms was automated through script. Below are the series of steps that were involved in creation of algorithms:

  1. Collect all historical visits to category/sub-category listing pages: includes showing product/SKU impressions and resulting clicks with respect to each visit. It can also be tagged to “add to cart” or “purchase” transaction through listing page
  2. Define target flag: includes defining flags for modeling click/add to cart/purchase or a combination of these
  3. Create features against target flag: these features are based on product metadata, visitor engagement with category/sub-category (past click/purchase/add to cart of same SKU) etc. Typical attributes involve product specific features such as Brand, Color, Size, Pattern, Fit etc. and recency attributes. Product features help us discover visitor preferences with respect to each visitor whereas recency attributes help us access impact of previous visit in ranking
  4. Create algorithms at visitor level: use of regression algorithms (that are easier to explain and implement) to identify important attributes and related weights for prioritisation
  5. Share these attributes and associated weights with client’s technical team for implementation
Data preparation and creation of algorithms:
    1. Creation of data platform:

. setting up data storage and processing infrastructure on Azure
. ETL scripts to migrate datasets from Google Big Query/Solr/Relational DB

    1. Creation of analytical datasets for visitor at product category/sub-category level:

. ETL scripts to join datasets and create variables
. final processed dataset

    1. Creation of algorithms:

. automated scripts for creation of algorithms
. discover visitor’s product preferences and weights and insertion into relational/nosql db

Implementation of algorithms:
  1. share visitor preferences with client’s team and participate in integration
  2. deployment of ETL batch jobs for pulling data at desired frequency
Technical documentation & handover:
  1. share technical documentation with comprehensive description on creation of algorithms including retraining and updating scenario
  2. training client’s team members on retraining on azure platform
JOIN OUR COMMUNITY
I agree to have my personal information transfered to MailChimp ( more information )
Join over 3.000 like minded AI enthusiasts who are receiving our weekly newsletters talking about the latest development in AI, Machine Learning and other Automation Technologies
We hate spam. Your email address will not be sold or shared with anyone else.