Relational Data Augmentation for Natural Language Model Fine-Tuning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating sufficient amounts of training data in natural language form for language processing machine learning models is challenging when the available ground truth data is not in natural language form, particularly for contexts like relational databases.
Innovation Solution
A method involving retrieving raw data from a relational database, augmenting it with related information, and populating natural language response templates to automatically generate training data, which is then used to train the model through a supervised learning process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is stored in relational database format, then data organization and retrieval efficiency are improved, but the data cannot be directly used for training language processing models requiring natural language format
Solution Approach 1:
The patent introduces a data transformation layer that acts as an intermediary between the relational database and the language model training process. This intermediary converts structured database records into natural language format, enabling the bridge between structured data storage and unstructured data consumption requirements of language models.
Solution Approach 2:
The system changes the data representation parameters by transforming data from structured tabular format to natural language textual format. This parameter transformation allows the same underlying data to serve both database query purposes and language model training purposes without requiring duplicate storage.
2Reliability
If manual conversion of data to natural language format is performed, then data quality and relevance are improved, but the process is time-consuming and labor-intensive
Solution Approach 1:
The system implements self-service data transformation where the database structure and schema automatically guide the conversion process. The system leverages existing database metadata, table structures, and relationships to automatically generate natural language representations without requiring manual intervention or human annotators.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and structuring the data in the database before the actual transformation to natural language format. By organizing data with proper schemas, relationships, and metadata in advance, the system enables automated transformation without requiring manual preparation during the conversion process.
3Productivity
If sufficient training data is generated through automated processes, then training efficiency is improved, but the complexity of the data generation system increases
Solution Approach 1:
The system achieves multi-functionality by using a single unified transformation framework that handles various data types, table structures, and relationship configurations through a common paradigm. The same core mechanisms can transform different relational data structures into natural language without requiring separate specialized processes for each data type.
Data Source
AI summary
Aspects of the present disclosure provide techniques for training a language processing machine learning model. Embodiments include retrieving a set of raw data from a data store. Embodiments include populating, based on the set of data, a natural language response template that is associated with a sample natural language prompt. Embodiments include providing the sample natural language prompt and the set of raw data as training inputs to the language processing machine learning model. Embodiments include receiving a training output from the language processing machine learning model in response to the training inputs. Embodiments include adjusting one or more parameters of the language processing machine learning model based on comparing the training output to the populated natural language response template.