RAGChain Docs
  • Introduction
  • Quick Start
  • Installation
  • RAGchain Structure
    • File Loader
      • Dataset Loader
        • Ko-Strategy-QA Loader
      • Hwp Loader
      • Rust Hwp Loader
      • Win32 Hwp Loader
      • OCR
        • Nougat Loader
        • Mathpix Markdown Loader
        • Deepdoctection Loader
    • Text Spliter
      • Recursive Text Splitter
      • Markdown Header Splitter
      • HTML Header splitter
      • Code splitter
      • Token splitter
    • Retrieval
      • BM25 Retrieval
      • Hybrid Retrieval
      • Hyde Retrieval
      • VectorDB Retrieval
    • LLM
    • DB
      • MongoDB
      • Pickle DB
    • Reranker
      • BM25 Reranker
      • UPR Reranker
      • TART Reranker
      • MonoT5 Reranker
      • LLM Reranker
    • Benchmark
      • Auto Evaluator
      • Dataset Evaluators
        • Qasper
        • Ko-Strategy-QA
        • Strategy-QA
        • ms-marco
  • Utils
    • Query Decomposition
    • Evidence Extractor
    • Embedding
    • Slim Vector Store
      • Pinecone Slim
      • Chroma Slim
    • File Cache
    • Linker
      • Redis Linker
      • Dynamo Linker
      • Json Linker
    • REDE Search Detector
    • Semantic Clustering
  • Pipeline
    • BasicIngestPipeline
    • BasicRunPipeline
    • RerankRunPipeline
    • ViscondeRunPipeline
  • For Advanced RAG
    • Time-Aware RAG
    • Importance-Aware RAG
Powered by GitBook
On this page
  • Overview
  • Usage
  • Initialize
  • Extract
  1. Utils

Evidence Extractor

EvidenceExtractor Class Documentation

Overview

The EvidenceExtractor class is a utility that extracts relevant evidences from a list of passages based on a given question. It uses a Language Model (LLM) to identify and return the relevant fragments from the original passages.

Usage

Initialize

To use the EvidenceExtractor class, you first need to create an instance of the class.

from RAGchain.utils.evidence_extractor import EvidenceExtractor

extractor = EvidenceExtractor()

You can put additional parameter model_name and api_base for using custom model. Plus, you can put custom system prompt for extracting evidence. If not, default english prompt will be used.

Extract

After the EvidenceExtractor instance has been initialized, you can use the extract method to extract relevant evidences from a list of passages based on a given question. The extract method takes a question (a string) and a list of passages as input and returns the extracted relevant evidences.

question = "What is the atomic number of hydrogen?"
passages = [
    Passage(content="Hydrogen is the first atom in atomic table."),
    Passage(content="Is is lighter than air so it floats."),
    Passage(content="Hydrogen can cause fire.")
]
result = extractor.extract(question, passages)

The extract method returns a string containing the extracted relevant evidences. If there is no evidence related to the question in the passages, it returns 'No Fragment' if you are using default prompt.

PreviousQuery DecompositionNextEmbedding

Last updated 1 year ago