A Literature Review Architecture for Active Inference: Scalable Assertion Extraction, Nanopublications, and Citation-Weighted Hypothesis Scoring

Catalog Row110

Citation KeyFriedman2026LiteratureReviewArchitectureActive110

Paper FolderAvailable

DOI10.5281/zenodo.19461934

Overview

Extracted from the local paper documentation when available.

No prior automated system tracks hypothesis-level evidence across the full Active Inference and Free Energy Principle literature at scale. This work presents a living meta-analysis framework: literature is retrieved from arXiv, Semantic Scholar, and OpenAlex and deduplicated ( N = 819 in v2) via a canonical identifier hierarchy (DOI arXiv ID Semantic Scholar ID OpenAlex ID). Papers are classified into a three-tier taxonomy (A: core theory; B: tools and translation; C: application domains) across eight categories. An LLM-powered pipeline evaluates each abstract against eight core hypotheses, emitting structured nanopublications (directionality, confidence, reasoning) that populate an RDF-compatible knowledge graph scored by a citation-weighted evidence function. All extractions are machine-generated and not fully manually validated; hypothesis scores are preliminary and are most useful fo

Active InferenceFree Energy Principlemeta-analysisliving literature reviewnanopublicationsLLM extractionhypothesis scoringknowledge graphOpenAlexSemantic ScholararXiv

Use Notes

Concise findings and methods pulled from README/SKILL documentation.

Findings / Concepts

Multi-source retrieval, identifier-based deduplication, and a three-tier (A/B/C) taxonomy for mapping the field.
LLM-driven assertion extraction into nanopublications wired for RDF-style graphs and queryable evidence landscapes.
Citation-weighted scoring over eight hypotheses with explicit reporting of consensus vs. debate structure and automation caveats.
Open pipeline suitable for continuous (living) updates as new papers appear.

Methods / Techniques

Corpus construction from arXiv, Semantic Scholar, and OpenAlex; hierarchical deduplication of records.
Taxonomic labeling (eight categories under A/B/C); complementary topic modeling (e.g., NMF) and citation-network views.
LLM evaluation of abstracts against eight hypotheses; nanopublication records with directionality, confidence, and rationale.
Graph-level citation-weighted aggregation; interpretive emphasis on relative tiers and biases (publication bias, linguistic asymmetry) per manuscript discussion.
Reproducible implementation: https://github.com/ActiveInferenceInstitute/act inf metaanalysis

Citation

Plain-text citation for quick reuse.

Friedman, Daniel Ari. 2026. A Literature Review Architecture for Active Inference: Scalable Assertion Extraction, Nanopublications, and Citation-Weighted Hypothesis Scoring. Active Inference Journal.

Primary source Documentation BibTeX