7 min read

Experience AI-powered vector search in Empathy Platform

Integrating semantic knowledge in real use cases

Keyword search is limited to lots of situations, leading to frustration in shoppers. Lack of results diversity, failure in shoppers’ intent, and domain understanding are just some causes where the implementation of vector search and semantic models are key to go beyond keyword exact matching and offer shoppers a way to overcome some daily search issues to get meaningful results.

Empathy Platform leverages vector search to:


Curious about how Empathy Platform approaches vector search based on domain-constrained semantic models? Check out Semantic models in vector search.

Zero results

Escape the end of the road with vectors and avoid shoppers’ frustration when getting no results.

One of the biggest frustrating situations in search is when shoppers get zero results from their search query. They may feel like the search engine is not working properly, or that it doesn’t understand what they need. By leveraging vector search, you can help shoppers skip zero results by providing similar product recommendations according to your shoppers' preferences. The underlying vector system comes into play to return product results that are semantically similar to what shoppers are looking for when there’s no exact keyword match with your product catalogue.

Zero results use case for vector search

For example, the query “Something to wear for a party” brings no results since your catalog doesn’t have an exact keyword match. You use vector representations of products in your feed to find closely related query alternatives, such as “dress for a party” or simply “party”, which have similar product features or attributes and align with what shoppers want. Rather than finding a search roadblock, shoppers get meaningful product recommendations based on semantically similar alternatives, derived from a deeper understanding of their shopping intent.


Spell Check, Partial Results, and vector search based recommendations are all used as fallback mechanisms for zero results. However, Spell Check and Partial Results are based on keyword matching rather than semantic matching, so shoppers get a set of results that may not meet their expectations or main intent.

Query expansion

Enrich a set of results with vectors when getting low results.

The traditional keyword-based search may not always bring as many results as desired. Maybe shoppers are getting just 20 or 40 product results since their intent can’t be easily captured. In cases like this, the backend vector system can be called to expand the shopper search intention by considering semantic similarities between the search query and the information in your product catalogue.

Query expansion use case based on vector search

For example, after searching for the query “kimono”, the search brings just a few results. Based on semantic search term relations like “kaftan”, you offer a broader range of results with carousels of product recommendations that are highly relevant to your shoppers, increasing the chances to finding products that meet their needs.

Synonym suggestions

Leverage semantic similarity to get similar search term suggestions on the fly.

When you’re aware your product catalogue does not contain some search terms that shoppers mostly use for searching in your commerce store, you expand your search catalogue with query synonyms manually. You can leverage the use of semantic models to represent terms as vectors and get semantically similar search term suggestions to feed your catalogue, broadening the search results space.

Synonym suggestions use case based on vector search

Usually, you define query synonyms manually with the Synonymize tool in Empathy Platform Playboard. These synonyms feed your product to expand shoppers’ search query scope when there’s no exact keyword match. Empathy Platform leverages the power of vector search to faster define synonyms with semantically-related term suggestions that have been used by shoppers for searching in your commerce store. For example, for the query “sugar-free” you get synonyms suggestions from shoppers such as sugarless, unsugared, non sweet, sweetener, saccharine, or even calorie-free. So, the possibility of returning relevant product results to shoppers increase, even when the original search query doesn’t have an exact match in your product catalogue.


Want to know more about how Empathy Platform services communicate with the Semantics API to offer synonym suggestions? Check out the Synonym suggestions with Semantics API section in the Synonyms overview.

Semantic feed enrichment

Use semantic domain understanding to enrich the product catalogue.

Instead of using semantic knowledge for individual fallback search features on query time, vector-based mechanisms can be also implemented directly to the standard result set on index time to avoid potentially degraded search performance, maybe due to customers resource competition. Here, Empathy Platform leverages semantic models to feed the product catalogue with terms, keywords, and other vector search information that semantically enrich the catalogue data with real search information.

Semantic feed enrichment use case

Vector search helps you expand your product catalogue by finding similar data that is not explicitly listed in your feed, such as style, season, or even material-related attributes. Comparing the vector representation of products, items with similar attributes or characteristics can be identified. For example, for a basic spring & summer season catalogue that contains information about the product name, description, categories, and colors, you can enrich your catalogue with attributes and values such as “cotton” or “silk” for materials and “party”, “evening”, “wedding”, or “classic” for style. This allows you to increase the variety of products available and broaden shoppers’ options.

Horizontal monetization indexing

Elevate sponsored product relevance while meeting shopping intent.

As monetization reshapes retail search dynamics, integrating sponsored products with organic results on the search engine results page (SERP) becomes a complex task. Monetization techniques can lead to result pages populated with noisy sponsored products, making it challenging to ensure highly relevant sponsored results aligned with shopping intent.

Empathy Platform leverages horizontal indexing and vector search to optimize quality and relevancy in monetized products, preventing irrelevant sponsored products on the SERP. Here, a specific horizontal monetization index pipeline couples with vector semantic similarity techniques, understanding shopper intent and seamlessly blending organic results with strategically positioned sponsored products. Sponsored results not only capture shoppers' attention but aligns with their shopping and discovery journey, maximizing the engagement and conversion potential sought by retailers and CPGs.

Monetization blended with vector search use case

With a specific index pipeline for monetization, new ranking signals enrich the product catalogue, leading to higher search relevance. You can fine-tune ranking configuration settings in the Empathy Platform’s Playboard to improve the relevance of both monetization and vector search attributes, finding the ideal balance between semantic matching and sponsored products' relevancy. For example, a search query like “tomato” returns a mix of organic tomato options and brand-sponsored tomato sauce products placed in the top positions of the SERP. These monetized results relate to and complement the search intent, inspiring your shoppers with context-aware sponsored products.


Want to know more on how to leverage horizontal indexing and vector search for monetized-based index? Check out Vector and keyword search as a unified index.

Combine vectors and design for a new content search experience.

When understanding certain catalogue architectures, like content-based catalogues, becomes a roadblock for search engines, semantic knowledge is key to expressing content category relationships and relevancy. Using semantic models helps content clustering by establishing semantic relationships between content, documents, and topics.

Traditionally, relevancy in search has been primarily determined by keyword scoring and the position of results on the results grid. However, this conventional view falls short when representing content as vectors. It's when the holon-based experience comes into play as a new visual way of expressing content relevancy and categorization through color and size, where you visually navigate into categories that lead you to new category paths and final doc results.

vectorized content search use case

Enabling a holon discovery experience based on vectors enriches the content discovery journey by providing a visual structure where size indicates relevancy and colors show content categories. These categories derive from semantic similarity, ensuring a domain-constrained and privacy-oriented approach. For example, for the query "search" multiple holon results display showing topic-related final results (content pages) and categories. By clicking a category holon, you're taken deeper into specific search results.

Try it out!

Try yourself the holon-based discovery experience by searching for a topic in the Empathy Platform Docs search bar right away.