Maestro: Automatic Generation of Comprehensive Benchmarks for Question Answering Over Knowledge Graphs
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
Recently, there has been an upsurge in the number of knowledge graphs (KG) that can only be accessed by experts. Non-expert users lack an adequate understanding of the queried knowledge graph's vocabulary and structure, as well as the syntax of the structured query language used to express the user's information needs. To increase the user base of these KGs, a set of Question Answering (QA) systems that use natural language to query these knowledge graphs have been introduced. However, finding a benchmark that accurately evaluates the quality of a QA system is a difficult task due to (1) the high degree of variation in the fine-grained properties among the existing benchmarks, (2) the static nature of the existing benchmarks versus the evolving nature of KGs, and (3) the limited number of KGs targeted by existing benchmarks, which hinders the usability of QA systems in real-world deployment over KGs that are different from those that were used in the evaluation of the QA systems. In this paper, we introduce Maestro, a benchmark generation system for question answering over knowledge graphs. Maestro can generate a new benchmark for any KG given the KG and, optionally, a text corpus that covers this KG. The benchmark generated by Maestro is guaranteed to cover all the properties of the natural language questions and queries that were encountered in the literature as long as the targeted KG includes these properties. Maestro also generates high-quality natural language questions with various utterances that are on par with manually-generated ones to better evaluate QA systems.
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 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.004 | 0.004 |
| 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