Multi-step Iterative Automated Domain Modeling with Large Language Models
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Domain modeling, which represents the concepts and relationships in a problem domain, is an essential part of software engineering. As large language models (LLMs) have recently exhibited remarkable ability in language understanding and generation, many approaches are designed to automate domain modeling with LLMs. However, these approaches usually formulate all input information to the LLM in a single step. Our previous single-step approach resulted in many missing modeling elements and advanced patterns. This paper introduces a novel framework designed to enhance fully automated domain model generation. The proposed multi-step automated domain modeling approach extracts model elements (e.g., classes, attributes, and relationships) from problem descriptions. The approach includes instructions and human knowledge in each step and uses an iterative process to identify complex patterns, repeatedly extracting the pattern from various instances and then synthesizing these extractions into a summarized overview. Furthermore, the framework incorporates a self-reflection mechanism. This mechanism assesses each generated model element, offering self-feedback for necessary modifications or removals, and integrates the domain model with the generated self-feedback. The proposed approach is assessed in experiments, comparing it with a baseline single-step approach from our earlier work. Experiments demonstrate a significant improvement over our earlier work, with a 22.71% increase in the F1-score for identifying classes, 75.18% for relationships, and a 10.39% improvement for identifying the player-role pattern, with comparable performance for attributes. Our approach, dataset, and evaluation provide valuable insight for future research in automated LLM-based domain modeling.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it