Microsoft GraphRAG 분석 및 최적화된 GraphRAG를 활용하기 위한 노하우 및 사용법

GraphRAG auto-tuning provides rapid adaptation to new domains

Manually creating and tuning a set of domain-specific prompts is time-consuming. We know, as all the prompts used for news articles were generated manually. To streamline this process, we developed an automated tool that generates domain-specific prompts, which are tuned and ready to use. This tool follows a human-like approach; we provided an LLM with a sample of text data (e.g., 1% of 10,000 chemistry papers) and instructed it to produce the prompts it deemed most applicable to the content. Now, with these automatically generated and tuned prompts, we can immediately apply GraphRAG to a new domain of our choosing, confident that we’ll get high-quality results.

During the indexing process, GraphRAG uses a set of prompts to instruct the LLM as it reads through the source content, extracting and organizing relevant information to construct the knowledge graph. Three of GraphRAG’s main indexing prompts include:

1. Entity and relationship extraction: Identifies all the entities present and establishes relationships among them.

2. Entity and relationship summarization: Consolidates instances of entities and their relationships into a single, concise description.

3. Community report generation: Generates a summary report for each community within the constructed knowledge graph.

• Extraction instructions: Provide the LLM with guidance on how to perform extraction.

• Few-shot examples: Supply the LLM real examples of the types of entities and relationships worth extracting.

• Real data: Serves as a placeholder that is replaced by chunks of source content.

• Gleanings: Encourage the LLM, over multiple turns, to extract additional information.

**Goal**
_Given a text document that is potentially relevant to this activity and a list of entity types, identify all entities of those types from the text and all relationships among the identified entities._

**Steps**

1. **Identify all entities.** For each identified entity, extract the following information:

- _entity_name_: Name of the entity, capitalized
- _entity_type_: One of the following types: [{entity_types}]
- _entity_description_: Comprehensive description of the entity’s attributes and activities

Format each entity as (“entity“, , , )

12. **From the entities identified in step 1, identify all pairs of (_source_entity_, _target_entity_) that are ***clearly related*** to each other.**

- _source_entity_: name of the source entity, as identified in step 1
- _target_entity_: name of the target entity, as identified in step 1
- _relationship_description_: explanation as to why you think the source entity and the target entity are related to each other
- _relationship_strength_: a numeric score indicating strength of the relationship between the source entity and target entity

Format each relationship as (“relationship“, , , , )

25. **Return output in English as a single list of all the entities and relationships identified in steps 1 and 2.** Use **{record_delimiter}** as the list delimiter.

27. **When finished, output **{completion_delimiter}****

###################### **Examples** ######################

**Example 1**:
_Entity_types_: ORGANIZATION,PERSON
_Text_:
_The Verdantis’s Central Institution is scheduled to meet on Monday and Thursday, with the institution planning to release its latest policy decision on Thursday at 1:30 p.m. PDT, followed by a press conference where Central Institution Chair Martin Smith will take questions. Investors expect the Market Strategy Committee to hold its benchmark interest rate steady in a range of 3.5%-3.75%._

**Output**:
(“entity”, **CENTRAL INSTITUTION**, ORGANIZATION, _The Central Institution is the Federal Reserve of Verdantis, which is setting interest rates on Monday and Thursday_)
(“entity”, **MARTIN SMITH**, PERSON, _Martin Smith is the chair of the Central Institution_)
(“entity”, **MARKET STRATEGY COMMITTEE**, ORGANIZATION, _The Central Institution committee makes key decisions about interest rates and the growth of Verdantis’s money supply_)

(“relationship”, **MARTIN SMITH** – **CENTRAL INSTITUTION**, _Martin Smith is the Chair of the Central Institution and will answer questions at a press conference_, 9)

**Example 2**:
_Entity_types_: ORGANIZATION
_Text_:
_TechGlobal’s (TG) stock skyrocketed in its opening day on the Global Exchange Thursday. But IPO experts warn that the semiconductor corporation’s debut on the public markets isn’t indicative of how other newly listed companies may perform._
_TechGlobal, a formerly public company, was taken private by Vision Holdings in 2014. The well-established chip designer says it powers 85% of premium smartphones._

**Output**:
(“entity”, **TECHGLOBAL**, ORGANIZATION, _TechGlobal is a stock now listed on the Global Exchange which powers 85% of premium smartphones_)
(“entity”, **VISION HOLDINGS**, ORGANIZATION, _Vision Holdings is a firm that previously owned TechGlobal_)

(“relationship”, **TECHGLOBAL** – **VISION HOLDINGS**, _Vision Holdings formerly owned TechGlobal from 2014 until present_, 5)

**Example 3**:
_Entity_types_: ORGANIZATION,GEO,PERSON
_Text_:
_Five Aurelians jailed for 8 years in Firuzabad and widely regarded as hostages are on their way home to Aurelia._
_The swap orchestrated by Quintara was finalized when $8bn of Firuzi funds were transferred to financial institutions in Krohaara, the capital of Quintara._
_The exchange initiated in Firuzabad’s capital, Tiruzia, led to the four men and one woman, who are also Firuzi nationals, boarding a chartered flight to Krohaara._
_They were welcomed by senior Aurelian officials and are now on their way to Aurelia’s capital, Cashion._
_The Aurelians include 39-year-old businessman Samuel Namara, who has been held in Tiruzia’s Alhamia Prison, as well as journalist Durke Bataglani, 59, and environmentalist Meggie Tazbah, 53, who also holds Bratinas nationality._

**Output**:
(“entity”, **FIRUZABAD**, GEO, _Firuzabad held Aurelians as hostages_)
(“entity”, **AURELIA**, GEO, _Country seeking to release hostages_)
(“entity”, **QUINTARA**, GEO, _Country that negotiated a swap of money in exchange for hostages_)
(“entity”, **TIRUZIA**, GEO, _Capital of Firuzabad where the Aurelians were being held_)
(“entity”, **KROHAARA**, GEO, _Capital city in Quintara_)
(“entity”, **CASHION**, GEO, _Capital city in Aurelia_)
(“entity”, **SAMUEL NAMARA**, PERSON, _Aurelian who spent time in Tiruzia’s Alhamia Prison_)
(“entity”, **ALHAMIA PRISON**, GEO, _Prison in Tiruzia_)
(“entity”, **DURKE BATAGLANI**, PERSON, _Aurelian journalist who was held hostage_)
(“entity”, **MEGGIE TAZBAH**, PERSON, _Bratinas national and environmentalist who was held hostage_)

(“relationship”, **FIRUZABAD** – **AURELIA**, _Firuzabad negotiated a hostage exchange with Aurelia_, 2)
(“relationship”, **QUINTARA** – **AURELIA**, _Quintara brokered the hostage exchange between Firuzabad and Aurelia_, 2)
(“relationship”, **QUINTARA** – **FIRUZABAD**, _Quintara brokered the hostage exchange between Firuzabad and Aurelia_, 2)
(“relationship”, **SAMUEL NAMARA** – **ALHAMIA PRISON**, _Samuel Namara was a prisoner at Alhamia prison_, 8)
(“relationship”, **SAMUEL NAMARA** – **MEGGIE TAZBAH**, _Samuel Namara and Meggie Tazbah were exchanged in the same hostage release_, 2)
(“relationship”, **SAMUEL NAMARA** – **DURKE BATAGLANI**, _Samuel Namara and Durke Bataglani were exchanged in the same hostage release_, 2)
(“relationship”, **MEGGIE TAZBAH** – **DURKE BATAGLANI**, _Meggie Tazbah and Durke Bataglani were exchanged in the same hostage release_, 2)
(“relationship”, **SAMUEL NAMARA** – **FIRUZABAD**, _Samuel Namara was a hostage in Firuzabad_, 2)
(“relationship”, **MEGGIE TAZBAH** – **FIRUZABAD**, _Meggie Tazbah was a hostage in Firuzabad_, 2)
(“relationship”, **DURKE BATAGLANI** – **FIRUZABAD**, _Durke Bataglani was a hostage in Firuzabad_, 2)

######################
**Real Data**
######################

**Entity_types**: {entity_types}
**Text**: {input_text}

**Output**:

Introducing DRIFT Search: Combining global and local search methods to improve quality and efficiency

DRIFT Search: A step-by-step process

GraphRAG: Improving global search via dynamic community selection

LazyGraphRAG: Setting a new standard for quality and cost

GraphRAG

Indexing

Prompt tuning

• Entity/Relationship Extraction

• Entity/Relationship Description Summarization

• Claim Extraction

• Community Reports

Query engine

Local Search

Global Search

Architecture

Dataflow

CLI로 사용하기

Index

Prompt-tune

Query

Auto tuning

Document Selection Methods

• random: Select text units randomly. This is the default and recommended option.

• top: Select the head n text units.

• all: Use all text units for the generation. Use only with small datasets; this option is not usually recommended.

• auto: Embed text units in a lower-dimensional space and select the k nearest neighbors to the centroid. This is useful when you have a large dataset and want to select a representative sample.

Manual tuning

끝으로, 주목 및 개선해보면 좋을 부분