Skip to content

v0.3.1 - Hybrid Entity Resolution

Choose a tag to compare

View all tags
@junhewk junhewk released this 29 Jan 01:48
· 64 commits to master since this release
0.3.1
f70fcd7

What's New

Hybrid Entity Resolution (Opt-in)

A multi-stage entity deduplication pipeline inspired by recent knowledge graph research (KGGen, NeurIPS 2025). Ensures "AI", "artificial intelligence", and "Artificial Intelligence" merge into a single node.

Resolution Pipeline:

Stage Method Speed
1. Persistent cache Previously resolved tokens O(1)
2. Session cache Same name resolved this session O(1)
3. Exact name Hash lookup on canonical name O(1)
4. Alias match Hash lookup on stored aliases O(1)
5. Embedding similarity Cosine similarity > 0.90 = auto-merge O(n)
6. LLM verification Ambiguous matches (0.80-0.90) verified by LLM API call
7. Create new No match found -

New Features

  • Embedding-based resolution: Support for OpenAI, Gemini, and Ollama embedding providers
  • Binary embedding storage: Efficient embeddings.bin format (~6KB per node)
  • Manual entity merge: Right-click nodes in graph view to merge duplicates
  • Entity Resolution settings: Configure thresholds, providers, and cache management

Settings

  • Enable/disable embedding-based resolution (default: off to avoid API costs)
  • Choose embedding provider (can differ from main LLM provider)
  • Configure similarity thresholds (high: 0.90, low: 0.80)
  • Enable/disable LLM verification for ambiguous matches
  • Compute embeddings for existing nodes
  • Clear resolution cache

Documentation

  • Updated README with entity resolution docs
  • Added research references (LightRAG, Microsoft GraphRAG, KGGen, Neo4j GraphRAG, pgvector)

Installation

Using BRAT (Recommended)

  1. Install BRAT from Community Plugins
  2. Open command palette → "BRAT: Add a beta plugin"
  3. Enter: junhewk/simple-graph-builder

Manual

Download main.js, styles.css, and manifest.json below and place in .obsidian/plugins/simple-graph-builder/

Assets 5
Loading