Improving Search Results for Short- and Long-Tail Queries

Improving Search Results for Short- and Long-Tail Queries

This post was written by Alexander Golubev, machine learning engineer at Constructor.

In the dynamic world of ecommerce, understanding user search patterns is fundamental. The thorough analysis of user behavior underlines a critical insight: the majority of conversions/purchases is dominated by the most popular queries. 

However, the role of long-tail queries cannot be neglected. They enrich the overall user experience, lend credibility to the platform’s breadth, assist in finding niche items, and shape users’ perception of search efficiency.

While optimizing the entirety of user searches is essential for holistic growth, a concentrated focus on the most impactful queries provides a direct avenue to substantial business gains.

To be able to optimize for both, first of all it’s crucial to be able to measure your search performance on different segments. The awareness of how X% lift from an algorithm is distributed across query segments leads to a better understanding of quality expectations and potential directions for improvement. Besides, these queries in most cases should be treated differently, maybe even with different algorithms.

However, in terms of business impact, the real value predominantly lies in those top-performing queries. Modern ecommerce platforms are increasingly using machine learning to improve search capabilities. These models must be meticulously optimized for the most popular queries since even a 1% enhancement in these can lead to a significant increase in revenue.

Engineers and data scientists should focus on training ML models to recognize the importance of these queries and ensure precision and speed in delivering results. These models should extract the user’s intent (e.g. black pullover for women with striped pattern), decompose to more granular queries (e.g. clothes for the active weekend), or identify and correct ambiguities (e.g. reliable sneakers for rain). Marketing and sales teams, meanwhile, can leverage insights from top queries by coordinating promotions, inventory strategies, and product recommendations based on the data from these top-performing search results. 

At Constructor, we prioritize perfection in our search results, especially for the most popular queries. Even when results seem relevant, we believe there’s always room for improvement. 

But we don’t underestimate the significance of long-tail queries. While they may not dominate our sales, they are essential for offering a comprehensive shopping experience. Having a fully integrated ML platform in search which is being optimized separately for both long-tail and popular queries helps to address different segments and leads to great overall performance. By continually refining our approaches to these types of queries, we ensure we remain at the forefront of ecommerce best practices and provide the best possible service to our customers.

short- and long-tail queries