README.md

metadata

license: apache-2.0
language:
  - en
tags:
  - RAG
  - Embeddings
  - Clustering
  - Graph
  - knowledge graph
  - GraphRAG
size_categories:
  - 1M<n<10M

Dataset for Benchmarking RAG, GraphRAG and LLM Embeddings

Benchmarking LLM embedding is still based on comparing it with another LLM or QA datasets.

This dataset offers another method to test or compare the LLM embeddings.

This all-against-all comparison of text data Wikipedia Embeddings dataset

As the original dataset contains embeddings from three different LLMs, this can be programmatically checked and manually verified.

RAG systems can be tested to verify if the setup works as per the requirements. Because this data contains both embeddings and actual clustered data, manual verification is also easier.

Graph-RAG now uses extraction of entities to make connections between documents. That removes the context from data. This dataset has actual lines/sentences/chunks compared in the dataset. This offers contextual links between documents where similar chunks can be connected based how much those chunks are similar.

Method

1. The actual data is fed to the Graph AI method after being split into sentences. These are linguistic sentence structures.
2. This makes each entry in the original dataset as documents.
3. The Graph AI compares the sentences method similar to all-against-all comparison.
4. Resulting clusters of Sentences, Documents are then normalized for generating document graphs.

Uses

• Can be used for benchmark RAG, GraphRAG and Embeddings.
• Can be used for Vector based Clustering methods.

Schema

All files are new line JSON files. It could be loaded as JSONL

1. data_file.json - Base Data file

• doc_id - Unique ID from embedding files from original dataset. id === doc_id
• data - data from embedding files from original dataset. body == data

2. line_data.json - Contains lines used for Clustering

• doc_id - Unique ID
• line_id - Unique ID
• line - Splited chunk of line from data.

3. rows_clusters.json

• cluster_id - Unique ID denoting cluster.
• cluster - List of line_id that forms a conceptual cluster.
• confidence - Score of confidence showing how much these line_id are related.

4. docs_clusers.json

• cluster_id - Unique ID denoting cluster
• cluster - List of doc_id that forms a conceptual cluster.
• confidence - Score of confidence showing how much these doc_id are related.

5. doc_graph.json

• doc_id_1 - Unique ID
• doc_id_2 - Unique ID
• confidence - Score of confidence showing how much these two doc_id are related.
• lines - List of line_id that connects these two doc_id