New neural network-based recommendation algorithm shows relevant products up to 23% more often

We have developed a search algorithm based on neural networks. Compared to our previous algorithm, the new algorithm shows relevant products up to 23% more often and up to 14% more accurately predicts the next product a user will interact with. We tell you about the other benefits in the article.

Key benefits of the new algorithm

Works better on new or rare search phrases

User searches for “trellis with mirror”. The neural network “understands” the meaning of the search query and combines it with other phrases: “trellis”, “dressing table with mirror”, “dressing table”, “table with mirror”. This is how the algorithm builds the most relevant recommendations to a rare query.

For popular requests, it gives more precise recommendations

A user searches for “laptop for graphic design”. Instead of showing a list of all available laptops, the new algorithm will take the search query apart. It will understand the specific laptop needed and suggest options with high performance and a quality display.

An algorithm that doesn’t use neural networks would show the laptops that people were most likely to buy for a given query. The new algorithm suggests models that better solve the specific need.

Reduces the dependence of recommendations on product popularity.

In many online stores, the old algorithm too often showed super popular products, sometimes overshadowing more relevant options. This approach caused conversion rates to drop. The new algorithm suggests products that more closely match the buyer’s intent.

The dependence of recommendations on product popularity is also called the “banana problem”. Bananas are one of the most popular items in grocery stores, and they have a strong influence on almost all recommendation blocks. Even if a person is looking for ground beef, he is likely to see this popular fruit in the recommendations. The new algorithm knows how to solve this problem and completely removes irrelevant products.

How to connect 

Contact your personal manager, who will advise you on whether it is worth using the new algorithm in your case. And if you want to use neural networks in your work, he will help you set up a trial run. 

Leave a request for a demo to see the capabilities of the algorithm