🐦 Twitter Post Details

Viewing enriched Twitter post

@dair_ai

RAG systems struggle with multi-hop reasoning. In most cases, the problem isn't the LLMs. It's the retrieval system. Standard RAG treats each piece of evidence as equally reliable, ignoring how documents connect to each other. Why is this a problem? When questions require reasoning across multiple sources, single-shot retrieval often misses "bridge" documents whose entities aren't mentioned in the original query. Iterative retrieval helps, but it introduces new issues: LLM-guided graph traversal can hallucinate or become stuck on partial reasoning from previous steps. This new research introduces SA-RAG, a framework that applies spreading activation, a mechanism from cognitive psychology, to knowledge-graph-based retrieval. How does it work? Instead of relying on the LLM to decide which documents to fetch next, activation propagates automatically through a knowledge graph. Starting from entities matched to the query, activation spreads outward through weighted connections, with strength diminishing over distance. Documents linked to highly activated entities get retrieved. The system builds a hybrid structure during indexing. An LLM extracts entities and relationships from text chunks, creating a knowledge graph where documents connect to entities through "describes" links. At query time, seed entities are identified by embedding similarity, then activation flows through the graph in a breadth-first manner. On MuSiQue, SA-RAG alone achieves 67% answer correctness with phi4, outperforming naive RAG at 45% and CoT-based iterative retrieval at 55%. When combined with chain-of-thought iterative retrieval, it reaches 74% on MuSiQue and 87% on 2WikiMultiHopQA. This system demonstrates a 25% to 39% absolute improvement over naive RAG across benchmarks. Notably, these results come from small, open-weight models like phi4 and gemma3, which require no fine-tuning. Spreading activation captures associative relevance rather than surface-level similarity. The method works as a plug-and-play module, boosting any training-free RAG pipeline without architectural changes. Paper: https://t.co/jLZLkacDAX Learn to build effective RAG and AI agents in our academy: https://t.co/zQXQt0PMbG

Media 1
Media 2
View on Twitter

📊 Media Metadata

{
  "media": [
    {
      "type": "photo",
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2002426307389649084/media_0.png?",
      "filename": "media_0.png"
    },
    {
      "type": "photo",
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2002426307389649084/media_1.png?",
      "filename": "media_1.png"
    }
  ],
  "processed_at": "2025-12-20T18:09:06.359390",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "2002426307389649084",
  "url": "https://x.com/dair_ai/status/2002426307389649084",
  "twitterUrl": "https://twitter.com/dair_ai/status/2002426307389649084",
  "text": "RAG systems struggle with multi-hop reasoning.\n\nIn most cases, the problem isn't the LLMs. It's the retrieval system.\n\nStandard RAG treats each piece of evidence as equally reliable, ignoring how documents connect to each other.\n\nWhy is this a problem?\n\nWhen questions require reasoning across multiple sources, single-shot retrieval often misses \"bridge\" documents whose entities aren't mentioned in the original query. Iterative retrieval helps, but it introduces new issues: LLM-guided graph traversal can hallucinate or become stuck on partial reasoning from previous steps.\n\nThis new research introduces SA-RAG, a framework that applies spreading activation, a mechanism from cognitive psychology, to knowledge-graph-based retrieval.\n\nHow does it work?\n\nInstead of relying on the LLM to decide which documents to fetch next, activation propagates automatically through a knowledge graph. Starting from entities matched to the query, activation spreads outward through weighted connections, with strength diminishing over distance.\n\nDocuments linked to highly activated entities get retrieved.\n\nThe system builds a hybrid structure during indexing. An LLM extracts entities and relationships from text chunks, creating a knowledge graph where documents connect to entities through \"describes\" links. At query time, seed entities are identified by embedding similarity, then activation flows through the graph in a breadth-first manner.\n\nOn MuSiQue, SA-RAG alone achieves 67% answer correctness with phi4, outperforming naive RAG at 45% and CoT-based iterative retrieval at 55%. When combined with chain-of-thought iterative retrieval, it reaches 74% on MuSiQue and 87% on 2WikiMultiHopQA.\n\nThis system demonstrates a 25% to 39% absolute improvement over naive RAG across benchmarks. Notably, these results come from small, open-weight models like phi4 and gemma3, which require no fine-tuning.\n\nSpreading activation captures associative relevance rather than surface-level similarity. The method works as a plug-and-play module, boosting any training-free RAG pipeline without architectural changes.\n\nPaper: https://t.co/jLZLkacDAX\n\nLearn to build effective RAG and AI agents in our academy: https://t.co/zQXQt0PMbG",
  "source": "Twitter for iPhone",
  "retweetCount": 3,
  "replyCount": 0,
  "likeCount": 27,
  "quoteCount": 1,
  "viewCount": 4198,
  "createdAt": "Sat Dec 20 17:10:09 +0000 2025",
  "lang": "en",
  "bookmarkCount": 51,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2002426307389649084",
  "displayTextRange": [
    0,
    277
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "dair_ai",
    "url": "https://x.com/dair_ai",
    "twitterUrl": "https://twitter.com/dair_ai",
    "id": "889050642903293953",
    "name": "DAIR.AI",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1643277398522187778/31dedbLo_normal.jpg",
    "coverPicture": "https://pbs.twimg.com/profile_banners/889050642903293953/1742055232",
    "description": "",
    "location": "",
    "followers": 83612,
    "following": 1,
    "status": "",
    "canDm": true,
    "canMediaTag": true,
    "createdAt": "Sun Jul 23 09:12:45 +0000 2017",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 3920,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 94,
    "statusesCount": 2686,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "2002426307389649084"
    ],
    "profile_bio": {
      "description": "Democratizing AI research, education, and technologies.",
      "entities": {
        "description": {},
        "url": {
          "urls": [
            {
              "display_url": "dair.ai",
              "expanded_url": "https://www.dair.ai/",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/lkqPZtMmfU"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "display_url": "pic.twitter.com/G7texzvVa0",
        "expanded_url": "https://twitter.com/dair_ai/status/2002426307389649084/photo/1",
        "ext_media_availability": {
          "status": "Available"
        },
        "features": {
          "large": {
            "faces": [
              {
                "h": 63,
                "w": 63,
                "x": 360,
                "y": 1613
              }
            ]
          },
          "orig": {
            "faces": [
              {
                "h": 63,
                "w": 63,
                "x": 360,
                "y": 1613
              }
            ]
          }
        },
        "id_str": "2002426303149166593",
        "indices": [
          278,
          301
        ],
        "media_key": "3_2002426303149166593",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARvKDB0IGtABCgACG8oMHgTbcLwAAA==",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAABCgABG8oMHQga0AEKAAIbygweBNtwvAAA",
            "media_key": "3_2002426303149166593"
          }
        },
        "media_url_https": "https://pbs.twimg.com/media/G8oMHQga0AEOMFr.png",
        "original_info": {
          "focus_rects": [
            {
              "h": 905,
              "w": 1616,
              "x": 0,
              "y": 0
            },
            {
              "h": 1616,
              "w": 1616,
              "x": 0,
              "y": 0
            },
            {
              "h": 1792,
              "w": 1572,
              "x": 22,
              "y": 0
            },
            {
              "h": 1792,
              "w": 896,
              "x": 360,
              "y": 0
            },
            {
              "h": 1792,
              "w": 1616,
              "x": 0,
              "y": 0
            }
          ],
          "height": 1792,
          "width": 1616
        },
        "sizes": {
          "large": {
            "h": 1792,
            "w": 1616
          }
        },
        "type": "photo",
        "url": "https://t.co/G7texzvVa0"
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "urls": [
      {
        "display_url": "arxiv.org/abs/2512.15922",
        "expanded_url": "https://arxiv.org/abs/2512.15922",
        "indices": [
          2112,
          2135
        ],
        "url": "https://t.co/jLZLkacDAX"
      },
      {
        "display_url": "dair-ai.thinkific.com",
        "expanded_url": "https://dair-ai.thinkific.com/",
        "indices": [
          2196,
          2219
        ],
        "url": "https://t.co/zQXQt0PMbG"
      }
    ]
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "article": null
}