Relational Learning as World Modeling

arXiv 2026 34 method

TLDR

Introduces TabWM, a context-based relational world model that predicts action effects as structured edits and enables planning by imagining futures, trained on WorldRel-generated worlds.

Reasoning

The paper addresses a valuable gap by extending relational learning to action-driven dynamics, with a clear method and strong empirical gains on synthetic worlds. However, it lacks real-world validation and the abstract provides limited detail on baselines or scalability.

Read-first score

Read-first score 34, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 41.

Recency 5%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=result

Topical relevance 25%

58.6

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 315.

Keyword Scores

world model

world dynamics prediction

interactive world model

generative world model

world simulator

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Reframes relational learning as world modeling: predicting how a relational state changes under actions rather than only predicting from a static database state.
Introduces TabWM (Tabular World Model), which uses observed executions from an unseen relational world as context to predict sparse structured edits without parameter updates.
Introduces WorldRel, a generator of executable relational worlds for pretraining.
Uses explicit relational edit predictions to recursively imagine candidate futures and compare actions before selection.

Methodology

The paper generates executable relational worlds with WorldRel for pretraining and evaluates TabWM on 512 held-out worlds. TabWM conditions on observed executions from an unseen world to predict sparse structured edits, requiring correct operation, relational location, and value, without parameter updates. Predicted edits are applied recursively to imagine futures and support search over action sequences, evaluated by realized return.

Key Results

On 512 held-out worlds, complete-edit F1 rises from 0.016 with no observed executions to 0.685 with 64 observed executions. Broader search over predicted futures improves the realized return of selected action sequences.

Limitations

No explicit limitations are stated in the abstract.

Resources

Paper