Dataset for: SESR-Eval: Dataset for Evaluating LLMs in the Title-Abstract Screening of Systematic Reviews
Bibliographic record
Abstract
Introduction This is a dataset for: "SESR-Eval: Dataset to Evaluate LLMs in the Screening Process of Systematic Reviews". Folder structure data The `data`-folder contains: - Initial replication package selection (`1-replication-package-selection`) - Inter-rater reliablity agreement for replication package selection (`2-replication-package-selection-reliability-agreement`) - Processed replication packages (`3-processed-data`) - Replication packages are omitted due to size constraints, but are downloadable via provided links - LLM results (`4-llm-results`) - The SESR-Eval dataset (`sesr-eval-dataset`) See: `data/sesr-eval-dataset/README.md` documentation The `documentation`-folder contains miscellaneous documentation for the study. experiments The `experiments`-folder contains the LLM experiment source code. How to run the benchmarks? 1. Install Python 3 2. Run `python3 -m venv venv` 3. Run `source venv/bin/acticate` 4. Run `pip install -r requirements.txt` 5. Copy `.env.example` to `.env` 6. Obtain: 1. Dataset (see data/sesr-eval-dataset/README.md) 2. OpenAI API key 3. Openrouter API key (if you wish to run other models than OpenAI) 7. Run: `./run_experiments.sh` Requirements - Python 3 Scopus API usage The data was downloaded from Scopus API between January 1 and 18 July, 2025 via http://api.elsevier.com and http://www.scopus.com. License The replication package is licensed with the CC-BY-ND 4.0 license. Each dataset secondary study has their own license. However, Elsevier has their own terms and conditions regarding the use of our research data: ---- This work uses data that was downloaded from Scopus API between Jan 1 and Apr 24, 2025 via http://api.elsevier.com and http://www.scopus.com. Elsevier allows access to the Scopus APIs in support of academic research for researchers affiliated with a Scopus subscribing institution. The end product here is a scholarly published work, that utilizes publications in Scopus for our research effort. We want to publish a scholarly work regarding Scopus data relationships. The data downloaded from Scopus API, for our work, is published to make work follow the practices of open science. It also makes possible to reproduce our work's results. Elsevier allows this use case under the following conditions, which our work meets: - The research is for non-commercial, academic purposes only. - The research is performed by approved representative of the applying institution. - The research is limited to the scope of Software engineering (SE) - we are not mining the entire Scopus dataset. - The retention of original research dataset is limited to archival purposes and reproduction of the research results. - Public sharing of data for purpose of reproducibility with a specific party is permissible upon written request and explicit written approval. - Scopus has been identified as the data source as described in the Scopus Attribution Guide. - If the user is a bibliometrician doing work outside this use case, they contact Elsevier's International Center for the Study of Research. The data is not displayed in a website or in a public forum outisde of the output format of the scholarly published work. The data is only stored in Zenodo, in this replication package.
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.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.011 | 0.114 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.003 | 0.005 |
| Bibliometrics | 0.007 | 0.010 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.005 | 0.003 |
| Open science | 0.003 | 0.004 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.322 | 0.090 |
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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".