Convermax presents Hybrid Search
Convermax takes search technology to the next level and presents Hybrid Search. The new search engine considers contextual meaning and delivers more precise and relevant results. It ensures your shoppers find exactly what they’re looking for, even if they use different wording or unusual phrasing.
What is Hybrid Search?
Hybrid search is an advanced search technology that integrates two key methodologies, such as traditional keyword-based searching and AI-powered semantic search. Convermax utilizes the rapidly developed large-scale LLMs trained on enormous amounts of data to provide even more efficient product results.
For standard keyword search, each query and document equal sparse vectors where each dimension corresponds to a specific term. Such vectors mostly consist of zeros, and few non-zeros represent the precise words. It’s efficient for exact matches, but it might not return results when the synonyms are used.
If a customer uses a keyword search for "2021 Honda Civic brake pads", the sparse vector would look like this:
The search engine looks for product descriptions that exactly match the description of Brake pads for the Honda Civic 2021 and ranks results based on TF-IDF (Term Frequency-Inverse Document Frequency).
Term Frequency measures how often a word (e.g., “brakes”) appears in a document. Products with frequent occurrences of the query words appear higher in the results.
Inverse Document Frequency reduces the weight of common words (e.g., “parts”, “auto”) across multiple documents (products) while increasing the weight of more specific terms.
For semantic search each query and document equal dense vectors generated by AI models using techniques like word embeddings or contextual embeddings. Every vector carries a set of meanings, such as synonyms and different ways people phrase similar searches. It understands related concepts, but is more computationally intensive and may miss exact matches if semantic relevance is too broad.
If a shopper uses a semantic search for "braking system replacement parts", the dense vector would understand the semantic similarity between “braking system” and “brakes.” And even though the products might not have a “braking system” in their descriptions, the search returns the results of disc brake kits or brake pads.
The search engine gets scores from keyword and semantic searches, then the other AI model re-ranks both results and shows the final product results to your shoppers.
How does Convermax Hybrid Search work?
Convermax Hybrid Search leverages the techniques of both: keyword search and semantic search based on vector embeddings. The search engine combines the results retrieved using both methods and provides your shoppers with relevant results. Hybrid Search works in sync with Year-Make-Model selections and filters.
For instance, when a shopper searches for "truck tires," the Hybrid Search handles it perfectly and delivers what the user needs, even though products don’t contain any info that they are suited for trucks.
Another example: the store doesn’t offer a Spanish-language version and has no term "all-terrain" in product descriptions and attributes. Nevertheless, Hybrid Search still recognizes that “neumáticos todoterreno” refers to “all-terrain tires” and shows the relevant products.
It delivers more accurate results and reduces the chances of situations when the right product exists in your store, but an exact keyword search can’t show it to the potential customer.
It eliminates the need to manage synonyms, and stop words. It knows common terms and jargon and can process that without any manual configuration.
It understands other languages out of the box. Language support may vary, but it can handle many queries pretty well.
Limited availability
Currently, Hybrid Search is available only for stores with smaller inventories. We plan to gradually roll out this feature, starting with stores that have around 20,000 products and then extending it to larger inventories. Stores with around 1,000,000 products will require a dedicated setup.
More information on Hybrid search: