Hi Google Cloud Community,
I’m working on enhancing our retail search experience using Vertex AI Search for Retail and facing challenges with dynamic faceting. I’d like to share our current approach and seek your advice or suggestions on optimizing it further.
In our product catalog, we handle a wide range of categories. Facet keys that work well for one query (e.g., "bearings") may not be relevant for others (e.g., "shirts" or "food"). Static facet configurations or precomputed attributes often result in irrelevant or overly broad filtering options, negatively impacting the user experience. Also count of facets are over 4000, exceeds limit 200.
To address this, we’ve implemented a two-step search process:
Step 1: Retrieve Relevant Facet Keys
Step 2: Perform the Actual Product Search
While this approach improves relevance, it feels inefficient to perform two searches for every query. Additionally, configuring the dynamicFacetSpec in Vertex AI Search doesn’t seem to work as expected—facet keys are not generated dynamically based on the user query. Instead, we still get a predefined set of facet keys, which limits flexibility.
DynamicFacetSpec Usage:
How can I configure dynamicFacetSpec in Vertex AI Search to dynamically generate facet keys based on the query context?
Performance Optimization:
Are there ways to reduce the overhead of performing two searches while still maintaining the benefits of dynamic faceting? HOW TO DO IMPLEMENT THIS FEATURE WITH ONLY ONE SEARCH??? ( This is what I have been investigating... Please...)
Best Practices:
Are there any recommended practices or alternative approaches for handling dynamic faceting in a diverse product catalog?
Thank you so much.
User | Count |
---|---|
2 | |
1 | |
1 | |
1 | |
1 |