Retrieval Augmented Generation (RAG) Analysis

Retrieval-augmented generation (RAG) is a text analysis approach in which relevant passages are first retrieved from a document collection and then supplied to a language model to generate answers grounded in that evidence.

검색 증강 생성(RAG)은 문서 모음에서 관련 구절을 먼저 검색한 뒤 이를 언어모형에 제공하여 그 근거에 기반한 답변을 생성하는 텍스트 분석 방식이다.

How do I ask questions of a large set of documents and get answers grounded in the source text?

많은 문서에 질문하고 원문에 근거한 답을 얻으려면 어떻게 하나요?

Retrieval-augmented generation embeds documents into chunks, retrieves the passages most similar to a query, and has a language model answer using only those passages, so responses can cite their sources. It fits question answering and evidence extraction over reports, interviews or open-ended responses, but answers must be checked against the retrieved text.

When to use Retrieval Augmented Generation (RAG) Analysis

• Answering questions across many reports, transcripts or open-ended responses
• Extracting evidence with references to the source passages
• Document collections too large to read or fit into a single prompt

When to use something else

• You need the main themes across a corpus without specific questions: use topic modeling such as LDA
• You need a validated classification of every text: use supervised text classification

Assumptions and requirements

• A document collection split into retrievable chunks
• An embedding model and similarity search for retrieval
• Evaluation of retrieval relevance and answer faithfulness
• Human verification of generated answers

What the analysis reports

• Generated answers with cited source passages
• Retrieved passages and similarity scores
• Retrieval metrics (e.g., recall@k)
• Faithfulness or groundedness assessment

How to report it

Using retrieval-augmented generation over [n] documents, [value]% of generated answers were judged fully supported by the retrieved passages.

Related methods

• LDA Topic Modeling
• KeyBERT Keyword Extraction
• Cosine Similarity Analysis

Run Retrieval Augmented Generation (RAG) Analysis on Quorant