Product recommendations and similarity analysis are crucial for eCommerce, inventory management, and customer experience. Traditional methods often rely on rigid categorization or manual tagging, which can be limiting and unscalable. In this talk, we present an innovative approach using vector-based similarity to dynamically compare products in Odoo.

We’ll explore how explicit product characteristics (e.g., attributes, categories, specs) can be systematically encoded into binary vectors, enabling efficient similarity computation through distance metrics (e.g., Hamming, Jaccard). We’ll break down how different data types (categorical, numerical, selection) are transformed into vector components and aggregated into a unified representation.

Additionally, we’ll demonstrate how pre-trained semantic models (e.g., word embeddings, sentence transformers) can extend this system by converting product descriptions into dense vectors, capturing nuanced relationships beyond explicit features. In this presentation we will explain:
* The mathematical and practical foundations of vector similarity.
* How to implement this in Odoo taking advantage of the ts_vector extension for Postgresql and the new Vector field type we developed
* Performance trade-offs and optimizations for large-scale deployments.

This approach bridges structured and unstructured data, offering a flexible, scalable solution for recommendations, substitutions, and search enhancements—all while leveraging Odoo’s existing framework.

Target Audience:
Odoo developers, functional consultants, and technical architects looking to integrate document or folder structures with Odoo in a robust, scalable, and vendor-neutral way.