Ramgopal Prajapat:

Learnings and Views

E-commerce Product Ranking – Factors and Considerations

By: Ram on Aug 06, 2022

Ecommerce platforms have millions of products and showing the most relevant products to customers/users is pertinent. Machine Learning based Learning to Rank algorithms can help in displaying most relevant products for a given context.  Before start developing product ranking algorithm for e-commerce, the understanding the broader context and consideration is critical. In this blog, we will discuss the factors and dimensions involving in Product Ranking.

  • Context
  • Role Product Ranking for Ecommerce
  • Business Objective and Defining Relevance
  • Personalization in Product ranking
  • Additional Factors and Considerations
  • Concluding Thoughts

 

Context

Ecommerce platform such as Amazon and Flipkart in India have millions of products on their platform. For every need of customers, there are thousands of options available if not more. When I searched for “tshirts men”, there 4.6 million options (46,04,638 results for "tshirts men"). It shows 40 products on the first page and based on research not many customers go beyond the first page (according to Millward Brown reports, 70% of Amazon users do not go beyond the first page -clavisinsight.com 2015). Ecommerce platform should find a clever way to show the most relevant products to their customers and improve customer experience & conversion.

Like for search engine, where the documents that are most relevant are shown on the top, in Ecommerce, the most relevant products should be listed first in the order of relevance.

A family of Machine Learning algorithms called Learning to Rank are used to solve the product ranking problem for Ecommerce. In this blog, we will cover the approach and in the subsequent blog/s, we will develop the ranking model using open-source data and ML technique.

 

Role of Ranking in Ecommerce

The ecommerce is changing the way we do the shopping.  Two key journeys on ecommerce are search led (search the products of our interest) or exploration/discovery driven (click on brand or category of our need).  In both the journeys, for the selected search term of category/brand, there are hundreds and thousands of products available. But there is limited space on the screens to show the products.

Graphical user interface, website

Description automatically generated

Business Objective and Defining Relevance

In ecommerce, the ranking of product on the listing pages either post search or on the brand/category pages is ordering the products by their relevance. Now, we need to define relevance of products to the customers.  For example, if a customer has clicked on “Nike” brand and ‘Running shoes” category, we need to show the products that satisfy these conditions and are most relevant to the customer. There are hundreds of Nike Running shoe and we need to show the most relevant first.

There are multiple approaches to define most relevant products for either search query terms or brand/category pages.  And it can be defined in either of these two ways

  • Automated: Define Metric or set of metrics to define the relevance

    • CTR - when shown how many times the product is clicked, and it is defined as Click Through Rate. The products which are click more, probably customers are finding them relevant and hence they click, so show them on the top.
    • Conversions - Click through rate does not guarantee that the product was of real interest to customer unless customer places the order of the product. So, the products that are bought by a greater number of customers indicate that they are of higher relevance, hence show them first.

Advantages

    • Accurate – Products relevance to customers are defined by customer actions, so can be considered more accurate way to define their relevance to the customers/users. A similar approach is used by Google for organic search ranking.
    • Scalable - No dependency on the available of experts to rate the relevance for a given context.
    • Less costly – manual relevance ranking requires time and effort

Limitation

    • Cold start – for the new products, there will not be performance metrics and may not show up on the top. This can be challenging for ecommerce as the customers want to new products and designs.
  • Manual – group of expert assign relevance rating for the products for a category or brand

    • For each of the brand and category pages, the group of expert rates the products based on their relevance and assign the ranking
    • Use these expert ranking to develop ML based Learning to Rank models and assign relevance ranking to the other products that do not have ranking.

Advantages

    • Experts define the relevance and may be reflective of the business and context
    • No issue of cold start as they can be manually rated and handled

Limitation

    • Each of the expert may rate relevance differently
    • A lot of manual effort and dependent on their availability and not very scalable

Ranking and Personalization

  • Personalize the ranking of the products based on user data and context
  • Considering the indent of person (whether a customer is intent to purchase or explore), price points (whether a person is price concise or fashion forward) and brand affinity in the relevance ranking.

Additional Considerations

  • Sales and Discounts – how should ranking of the products be revised considering the sales period or level of discounts?  There may be good discounts on a brand, if these products are not visible on the category pages, there is loss of business revenue.
  • Seasons and Festivals – Relevance of product changes based on the seasons and festivals. We need to find a way to incorporate these into the relevance algorithms.
  • Handling New Products – how do you ensure new products are included in the list?
  • Product Diversity – How do we ensure that the products included in the ranking offer variety.

 

Concluding Thoughts

In this blog, we have discussed about the factors and dimensions to think before developing a product ranking algorithm for Online Ecommerce Platform. Here is a list of questions for you to share your perspective with me.

  • Do you come across any great example on what ecommerce platform learnt about your clicks and shown the products of your need?
  • Do you recollect any wow moment of ecommerce shopping? How could ecommerce know about this?
  • Do you remember any scenario where products shown were not great and you had to visit page 2, 3, etc.?

Looking forward to hear from you. Please comment with your questions and comments

Leave a comment