20 Jun 2023
From keywords to context: Exploring the potential of Intent Search
Yamine Durai
Yamine Durai
Marketing Manager

We are all chasing relevance. It's the linchpin that connects shoppers with the products they desire. However, traditional keyword-based search methods fall short of understanding buyer intent.

Let me explain. Let's say a shopper hits the query "red dress for a summer wedding" on an ecommerce website that relies solely on keyword-based search. The search algorithm will match the keywords "red," "dress," "summer," and "wedding" in the product listings (the stemming application could vary). In this case, the keyword-based search may retrieve a range of dresses, but it might not consider critical contextual factors. The search results could include red dresses for other seasons, summer dresses of different colors, and general white wedding gowns. Most of these results will be irrelevant. In fact, 31% of searches fetch zero results due to keyword-based search systems' limitations.

Now you might think that the solution is Vector Search. While that sounds like the utopia we want to be in, the intricacy of implementing Vector Search is a challenging ocean to cross.

So, how can your enterprise attain the golden relevancy that satisfies shoppers, generates revenue, and prevents lost opportunities?

Netcore Unbxd's Intent Search: An art mastered over years

Netcore Unbxd Intent Search combines Vector and keyword-based search models within a multi-dimensional Network. This approach leverages advanced machine learning and deep learning models to provide an efficient and effective solution for searching and retrieving contextually relevant information from vast datasets.

A precise, comprehensive, and scalable search experience

Let's start by understanding Vector Search, the foundation of Unbxd's Intent Search.

Vector Search is a semantic data retrieval technique that uses multi-dimensional (vector) spaces to plot and index unstructured data, including text, images, audio, and video. A multi-dimensional space is a conceptual space with many dimensions, each representing a specific attribute of the catalog items. It is a mathematical representation that allows for comparing and analyzing objects based on their various features.

An interesting fact is that Intent Search is the base for Visual Search, using which you can enable your shoppers to upload images as search queries and fetch visually similar products as a result.

That said, the fundamental assumption of Vector Search is that items with similar characteristics will have similar vectors. The closer the vectors are to each other, the more likely the items will be similar, based on algorithms such as cosine similarity and Euclidean distance. This way, your search can understand semantic relationships, enhancing search accuracy.

The story of finding an aviator-style sunglasses

Let's take an example of a sunglasses catalog. In Vector Search, each pair of sunglasses is represented as a vector. The dimensions of this space correspond to different characteristics or features of the sunglasses available in the catalog, like frame shape, lens color, frame material, and UV protection level.

Now, let's say a shopper wants to find sunglasses similar to a specific pair they have in mind. They provide a query with the characteristics of the desired sunglasses, such as "pilot style, black lenses, metal frame, and high UV protection." This search query is converted into a vector.

To fetch relevant results, the system ranks and retrieves sunglasses closest to the query vector based on the similarity score. This method ensures that the retrieved sunglasses have characteristics specified by the shopper.

What's even better is that the sunglasses retrieved in the search result might not have been explicitly tagged or labeled with the shopper's query. Despite the shopper not using predefined keywords, the system will offer accurate "aviator style sunglasses" as search results by leveraging semantic relationships encoded in vectors.

As mentioned above, the Vector Search process includes,

  1. Vector representation of the catalog data

  2. Plotting data points in the multi-dimensional space

  3. Indexing the vectorized values

  4. Query Processing

  5. Similarity comparison

  6. Fetching precise and relevant results

The limitations of Vector Search

While Vector Search offers several advantages, implementing it can be more complex than traditional keyword-based search. It requires preprocessing and indexing large catalogs to generate vector representations, which can be computationally intensive and time-consuming. Apart from that, Vector Search is sensitive to the size and dimensionality of the data. The efficiency and effectiveness can decrease as the dataset grows larger or becomes high-dimensional. Multi-dimensional data may suffer from the curse of dimensionality, where the distance between vectors becomes less meaningful.

Intent Search: The perfect union of Keyword-based and Vector Search

Intent search combines the power of vector search and keyword-based search to provide more accurate and personalized search results.

For example, when a shopper enters a query like "trendy sunglasses," the Intent Search system analyzes the query to understand the shopper's underlying intent. It extracts keywords from the query, such as "trendy" and "sunglasses," to capture the explicit preferences of the shopper.

Once the keywords are extracted, the Intent Search system utilizes a vector-based approach to comprehend the context and semantics of the query.

It maps the keywords to vectors, where each vector represents the characteristics and features of sunglasses. These vectors encode the relationships between styles, shapes, colors, and attributes. Intent Search identifies the most relevant vectors that align with the shopper's query by combining the keyword-based information with the Vector Space representation. It understands their semantic relationship and context. In this case, it seeks vectors that represent sunglasses with trendy styles. We can further enrich your catalog data with Generative AI to fill in the blanks seamlessly.

This approach enhances the search experience by providing highly relevant and personalized results that fit the shopper's requirements without going through the implementation challenges posed by Vector Search.

You can read more about Intent Search and deep dive into its functionalities in this whitepaper.

Achieving relevance in ecommerce is crucial for satisfying shoppers, driving revenue, and seizing opportunities. While traditional keyword-based search methods often fail to understand buyer intent, leveraging advanced technologies like Netcore Unbxd's Intent Search is the solution.

By adopting Intent Search, enterprises can unlock the golden relevancy that satisfies shoppers, generates revenue, and prevents lost opportunities. It empowers businesses to provide a seamless and tailored search experience, enabling shoppers to find exactly what they desire and enhancing their overall journey on your ecommerce site.