Systems and Methods for Creating and Continuously Improving Domain-Specific Translation Models Using Synthetic Data

A computer-implemented method for translating an arbitrary product name to a standardized product descriptor, the computer-implemented method comprising:

• receiving, by a domain-specific translation model, an arbitrary product name;
• generating, by the domain-specific translation model, a plausible standardized product descriptor corresponding to the arbitrary product name;
• retrieving, by the domain-specific translation model, a plurality of candidate standardized product descriptors from data storage based on the plausible standardized product descriptor;
• and
• selecting, by the domain-specific translation model, a candidate standardized product descriptor from the plurality of candidate standardized product descriptors.

A computer-implemented method for translating an arbitrary product name to a standardized product descriptor, the computer-implemented method comprising:

• receiving, by a domain-specific translation model, an arbitrary product name;
• generating, by the domain-specific translation model, a plausible standardized product descriptor corresponding to the arbitrary product name;
• retrieving, by the domain-specific translation model, a plurality of candidate standardized product descriptors from a canonical reference ontology (CRO) based on the plausible standardized product descriptor;
• performing, by the domain-specific translation model, a similarity search to compare each candidate standardized product descriptor against the arbitrary product name;
• and
• selecting, by the domain-specific translation model, a candidate standardized product descriptor from the plurality of candidate standardized product descriptors based on the similarity search.

A system for creating and continuously improving a domain-specific translation model, the system comprising:

• a canonical reference ontology (CRO) storing standardized product descriptors for a domain;
• an artificial intelligence (AI) model configured to generate synthetic training data based on the CRO;
• a domain-specific translation model configured to translate arbitrary product names to standardized product descriptors;
• and
• a feedback integration module configured to incorporate user corrections into the training corpus of the domain-specific translation model.

A computer-implemented method for continuously improving a domain-specific translation model, the computer-implemented method comprising:

• monitoring a canonical reference ontology (CRO) for new standardized product descriptors;
• detecting new standardized product descriptors;
• generating synthetic training data for the new descriptors using a generative artificial intelligence (GAI) model;
• collecting user feedback on translation results produced by a domain-specific translation model;
• combining the synthetic training data and the user feedback to create an updated training corpus;
• and
• retraining the domain-specific translation model with the updated training corpus.