{"work":{"id":"588618d7-fd41-4053-b34d-a981f8793039","openalex_id":null,"doi":null,"arxiv_id":"2404.16130","raw_key":null,"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","authors":null,"authors_text":"Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody","year":2024,"venue":"cs.CL","abstract":"The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as \"What are the main themes in the dataset?\", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these contrasting methods, we propose GraphRAG, a graph-based approach to question answering over private text corpora that scales with both the generality of user questions and the quantity of source text. Our approach uses an LLM to build a graph index in two stages: first, to derive an entity knowledge graph from the source documents, then to pregenerate community summaries for all groups of closely related entities. Given a question, each community summary is used to generate a partial response, before all partial responses are again summarized in a final response to the user. For a class of global sensemaking questions over datasets in the 1 million token range, we show that GraphRAG leads to substantial improvements over a conventional RAG baseline for both the comprehensiveness and diversity of generated answers.","external_url":"https://arxiv.org/abs/2404.16130","cited_by_count":null,"metadata_source":"pith","metadata_fetched_at":"2026-06-27T04:20:31.817638+00:00","pith_arxiv_id":"2404.16130","created_at":"2026-05-09T06:10:42.928972+00:00","updated_at":"2026-06-27T04:20:31.817638+00:00","title_quality_ok":true,"display_title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","render_title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization"},"hub":{"state":{"work_id":"588618d7-fd41-4053-b34d-a981f8793039","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":176,"external_cited_by_count":null,"distinct_field_count":13,"first_pith_cited_at":"2024-09-27T15:35:15+00:00","last_pith_cited_at":"2026-06-24T09:31:06+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-06-27T10:45:56.209237+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":39},{"context_role":"baseline","n":5},{"context_role":"method","n":2},{"context_role":"dataset","n":1}],"polarity_counts":[{"context_polarity":"background","n":38},{"context_polarity":"baseline","n":5},{"context_polarity":"use_method","n":2},{"context_polarity":"support","n":1},{"context_polarity":"use_dataset","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","claims":[{"claim_text":"The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as \"What are the main themes in the dataset?\", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these ","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"predicted scores ofnentities need to be supervised or ranked, the loss computation complexity is Closs(B) =O(Bn).(204) Therefore, when the selector is not enabled, the complexity of a single training batch is Ctrain =O(κ bw [Cenc(B) +C score(B) +C loss(B)]).(205) After enabling the query-conditioned subgraph selector, the training complexity becomes Csel train =O(κ bw [Cenc(B) +C score(B) +C loss(B) +C sel,train(B)]).(206) If a document-level loss is also explicitly added during training, then e","claim_type":"background","confidence":0.6,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks From Local to Global: A Graph RAG Approach to Query-Focused Summarization because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (1 contexts).","role_counts":[{"n":1,"context_role":"background"}]},"error":null,"updated_at":"2026-05-14T19:26:24.775606+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"5ba544d2-c775-4130-8639-2674734f3333","orcid":null,"display_name":"Darren Edge"},{"id":"a07e87e0-5f17-4fa4-a767-6eaaa3a2b639","orcid":null,"display_name":"Ha Trinh"},{"id":"a3506931-364c-4ccc-ac63-1805b615e126","orcid":null,"display_name":"Newman Cheng"},{"id":"863fdf02-c1a9-4240-bea2-1661ff77fb6e","orcid":null,"display_name":"Joshua Bradley"},{"id":"c450c5e5-1ab4-421a-9eba-7172bb9c3c67","orcid":null,"display_name":"Alex Chao"},{"id":"d36aca9b-32f1-40e4-9e98-19b128877400","orcid":null,"display_name":"Apurva Mody"}]},"error":null,"updated_at":"2026-05-14T19:26:26.472752+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T06:06:55.512725+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Retrieval-Augmented Generation for Large Language Models: A Survey","work_id":"b80d2790-6cd9-4c87-b3c4-de404f99a80e","shared_citers":28},{"title":"Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory","work_id":"a5aed26c-a248-48b6-a59e-f7693fcb180a","shared_citers":25},{"title":"Zep: A Temporal Knowledge Graph Architecture for Agent Memory","work_id":"515c933e-12ae-439d-a7ff-c07fee482dfb","shared_citers":20},{"title":"A-MEM: Agentic Memory for LLM Agents","work_id":"3b98feb2-fdb1-479a-bbe4-2c298a4592e2","shared_citers":19},{"title":"MemGPT: Towards LLMs as Operating Systems","work_id":"2698f5ad-c84c-40ca-b839-0912dae10ba2","shared_citers":19},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":19},{"title":"LightRAG: Simple and Fast Retrieval-Augmented Generation","work_id":"6118de04-b8eb-4163-826a-0f91a1bcdf14","shared_citers":17},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":12},{"title":"Voyager: An Open-Ended Embodied Agent with Large Language Models","work_id":"ffe0d207-86cf-4742-a100-e988ac8b9676","shared_citers":11},{"title":"Evaluating Very Long-Term Conversational Memory of LLM Agents","work_id":"2d8c9fb7-9ace-4925-8fe1-f4e44625d04c","shared_citers":10},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":10},{"title":"LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory","work_id":"9074870f-aee7-4103-b167-ac6473a8a9b3","shared_citers":10},{"title":"From rag to memory: Non-parametric continual learning for large language models","work_id":"f768fbda-cd1e-4a8e-b67c-040f4242fe93","shared_citers":9},{"title":"GPT-4o System Card","work_id":"f37bf1c7-4964-4e56-9762-d20da8d9009f","shared_citers":9},{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":9},{"title":"Corrective Retrieval Augmented Generation","work_id":"eda82a01-2f6c-4e23-9f7c-facff5b7e2ee","shared_citers":8},{"title":"Gemini: A Family of Highly Capable Multimodal Models","work_id":"83f7c85b-3f11-450f-ac0c-64d9745220b2","shared_citers":8},{"title":"arXiv preprint arXiv:2402.09727 , year=","work_id":"072fe48b-345d-4f4d-8de8-87adbfe72aa7","shared_citers":7},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":7},{"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","shared_citers":7},{"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","shared_citers":7},{"title":"Memory in the Age of AI Agents","work_id":"1ff75a14-302e-4906-aa86-1f96fbcf12ff","shared_citers":7},{"title":"Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning","work_id":"2a070440-2167-4398-be97-c2d4c3ee3541","shared_citers":7},{"title":"Mirix: Multi-agent memory system for llm-based agents","work_id":"bcb99885-6f76-4bd6-b442-e5f455bb6447","shared_citers":7}],"time_series":[{"n":1,"year":2024},{"n":2,"year":2025},{"n":96,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T06:06:41.049130+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T06:06:40.996603+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","claims":[{"claim_text":"The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as \"What are the main themes in the dataset?\", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these ","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"predicted scores ofnentities need to be supervised or ranked, the loss computation complexity is Closs(B) =O(Bn).(204) Therefore, when the selector is not enabled, the complexity of a single training batch is Ctrain =O(κ bw [Cenc(B) +C score(B) +C loss(B)]).(205) After enabling the query-conditioned subgraph selector, the training complexity becomes Csel train =O(κ bw [Cenc(B) +C score(B) +C loss(B) +C sel,train(B)]).(206) If a document-level loss is also explicitly added during training, then e","claim_type":"background","confidence":0.6,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks From Local to Global: A Graph RAG Approach to Query-Focused Summarization because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (1 contexts).","role_counts":[{"n":1,"context_role":"background"}]},"error":null,"updated_at":"2026-05-14T19:26:24.780509+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","claims":[{"claim_text":"The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as \"What are the main themes in the dataset?\", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these ","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks From Local to Global: A Graph RAG Approach to Query-Focused Summarization because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T06:06:48.568889+00:00"}},"summary":{"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","claims":[{"claim_text":"The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as \"What are the main themes in the dataset?\", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these ","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks From Local to Global: A Graph RAG Approach to Query-Focused Summarization because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Retrieval-Augmented Generation for Large Language Models: A Survey","work_id":"b80d2790-6cd9-4c87-b3c4-de404f99a80e","shared_citers":28},{"title":"Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory","work_id":"a5aed26c-a248-48b6-a59e-f7693fcb180a","shared_citers":25},{"title":"Zep: A Temporal Knowledge Graph Architecture for Agent Memory","work_id":"515c933e-12ae-439d-a7ff-c07fee482dfb","shared_citers":20},{"title":"A-MEM: Agentic Memory for LLM Agents","work_id":"3b98feb2-fdb1-479a-bbe4-2c298a4592e2","shared_citers":19},{"title":"MemGPT: Towards LLMs as Operating Systems","work_id":"2698f5ad-c84c-40ca-b839-0912dae10ba2","shared_citers":19},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":19},{"title":"LightRAG: Simple and Fast Retrieval-Augmented Generation","work_id":"6118de04-b8eb-4163-826a-0f91a1bcdf14","shared_citers":17},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":12},{"title":"Voyager: An Open-Ended Embodied Agent with Large Language Models","work_id":"ffe0d207-86cf-4742-a100-e988ac8b9676","shared_citers":11},{"title":"Evaluating Very Long-Term Conversational Memory of LLM Agents","work_id":"2d8c9fb7-9ace-4925-8fe1-f4e44625d04c","shared_citers":10},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":10},{"title":"LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory","work_id":"9074870f-aee7-4103-b167-ac6473a8a9b3","shared_citers":10},{"title":"From rag to memory: Non-parametric continual learning for large language models","work_id":"f768fbda-cd1e-4a8e-b67c-040f4242fe93","shared_citers":9},{"title":"GPT-4o System Card","work_id":"f37bf1c7-4964-4e56-9762-d20da8d9009f","shared_citers":9},{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":9},{"title":"Corrective Retrieval Augmented Generation","work_id":"eda82a01-2f6c-4e23-9f7c-facff5b7e2ee","shared_citers":8},{"title":"Gemini: A Family of Highly Capable Multimodal Models","work_id":"83f7c85b-3f11-450f-ac0c-64d9745220b2","shared_citers":8},{"title":"arXiv preprint arXiv:2402.09727 , year=","work_id":"072fe48b-345d-4f4d-8de8-87adbfe72aa7","shared_citers":7},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":7},{"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","shared_citers":7},{"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","shared_citers":7},{"title":"Memory in the Age of AI Agents","work_id":"1ff75a14-302e-4906-aa86-1f96fbcf12ff","shared_citers":7},{"title":"Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning","work_id":"2a070440-2167-4398-be97-c2d4c3ee3541","shared_citers":7},{"title":"Mirix: Multi-agent memory system for llm-based agents","work_id":"bcb99885-6f76-4bd6-b442-e5f455bb6447","shared_citers":7}],"time_series":[{"n":1,"year":2024},{"n":2,"year":2025},{"n":96,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"c450c5e5-1ab4-421a-9eba-7172bb9c3c67","orcid":null,"display_name":"Alex Chao","source":"manual","import_confidence":0.72},{"id":"d36aca9b-32f1-40e4-9e98-19b128877400","orcid":null,"display_name":"Apurva Mody","source":"manual","import_confidence":0.72},{"id":"5ba544d2-c775-4130-8639-2674734f3333","orcid":null,"display_name":"Darren Edge","source":"manual","import_confidence":0.72},{"id":"a07e87e0-5f17-4fa4-a767-6eaaa3a2b639","orcid":null,"display_name":"Ha Trinh","source":"manual","import_confidence":0.72},{"id":"863fdf02-c1a9-4240-bea2-1661ff77fb6e","orcid":null,"display_name":"Joshua Bradley","source":"manual","import_confidence":0.72},{"id":"a3506931-364c-4ccc-ac63-1805b615e126","orcid":null,"display_name":"Newman Cheng","source":"manual","import_confidence":0.72}]}}