Data-centric paradigm

As explained in Knowledge Graphs and LLMs in Action

A machine-learning development approach that prioritizes improving the quality and quantity of training data.

The data-centric paradigm improves machine-learning systems primarily through data engineering: increasing training-data quality and quantity instead of concentrating only on model architecture or hyperparameter tuning. It treats faulty data as a direct cause of faulty predictions and can reduce the need for model engineering in high-complexity models. section 1.2 section 5.3

A practical data-centric workflow emphasizes:

• Improve the training corpus before attempting further model changes, addressing both coverage and data quality. section 1.2
• Use natural-language task formulation and prompt engineering when capable LLMs make extensive model engineering unnecessary. section 5.3
• Validate outputs carefully: improved data does not eliminate hallucinations or unsupported reasoning when relevant information is absent from training data. section 5.3

Figure 5.3 Mental model of transforming textual data into structured knowledge: NER and RE using an LLM

Use the mental model to connect data-centric preparation with extracting named entities, relations, and properties from unstructured text into structured knowledge. section 5.3

How Knowledge Graphs and LLMs in Action covers it

How deep it goes

Defines the model-centric contrast and improvement targets, but stops short of prescribing detailed data-cleaning procedures.

Examples

Uses LLM-based extraction of named entities, relations, and properties from unstructured text, plus an unsupported rocket-cost answer.

What it assumes

Assumes traditional machine-learning concepts and teaches data quality and quantity as first-class development targets.

What you can do after

Apply it via data engineering to improve training corpora and formulate tasks for capable LLMs with less model engineering.

How it connects

Connects data quality with model engineering, prompt engineering, large language models, hallucinations, and knowledge graphs.

liveBookmanning.com

Read Knowledge Graphs and LLMs in Action on liveBook Get the book at manning.com

Related concepts

• data engineeringuses · Data engineering is the discipline of designing and operating systems and processes that collect, transform, integrate, store, and deliver data for computational use.

This explainer was generated by Manning's knowledge graph from the text of Knowledge Graphs and LLMs in Action. The section links point to the parts of the book it draws on.