Domain-Filtered Knowledge Graphs from Sparse Autoencoder Features

2026-04-26

AI Summary

This work addresses the limitations of existing sparse autoencoders, whose extracted features—though rich—are structurally flat and lack semantic coherence or domain focus, thereby failing to reveal the internal knowledge of language models. The authors propose a novel approach that constructs domain-specific concept sets through contrastive activation and multi-stage filtering, then builds co-occurrence graphs and cross-layer transcoder mechanism graphs, augmented with automatically annotated edges to produce human-readable knowledge graphs. This is the first method to transform sparse features into structured, multi-granular knowledge graphs enriched with semantic relationships, enabling a leap from local interpretability to global knowledge mapping. Evaluated on biology textbook data, the approach successfully reconstructs chapter-level knowledge structures, identifies cross-topic bridging concepts, and compresses thousands of sentence-level feature dimensions into compact, interpretable local knowledge views.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Knowledge Representation and Reasoning: Knowledge Acquisition

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Sparse autoencoders (SAEs) extract millions of interpretable features from a language model, but flat feature inventories aren't very useful on their own. Domain concepts get mixed with generic and weakly grounded features, while related ideas are scattered across many units, and there's no way to understand relationships between features. We address this by first constructing a strict domain-specific concept universe from a large SAE inventory using contrastive activations and a multi-stage filtering process. Next, we build two aligned graph views on the filtered set: a co-occurrence graph for corpus-level conceptual structure, organized at multiple levels of granularity, and a transcoder-based mechanism graph that links source-layer and target-layer features through sparse latent pathways. Automated edge labeling then turns these graph views into readable knowledge graphs rather than unlabeled layouts. In a case study on a biology textbook, these graphs recover coherent chapter and subchapter-level structure, reveal concepts that bridge neighboring topics, and transform messy sentence-level activity containing thousands of features into compact, readable views that illustrate the model's local activity. Taken together, this reframes a flat SAE inventory as an internal knowledge graph that converts feature-level interpretability into a global map of model knowledge and enables audits of reasoning faithfulness.

Problem

• sparse autoencoders
• feature interpretability
• knowledge graph
• domain concepts
• feature relationships

Innovation

• sparse autoencoders
• knowledge graphs
• domain filtering
• co-occurrence graph
• transcoder-based mechanism