Practical overview
When it fits
Use it when you want to explore COVID-19 research abstracts through semantic queries in a local, reproducible workflow, or when working with the repository’s model training scripts.
Who it is for
- Researchers seeking relevant COVID-19 and SARS-CoV-2 papers by querying their abstracts.
- Practitioners exploring sentence embeddings and transformer models for scientific-text search.
- Developers reproducing the repository’s documented NLI fine-tuning and STS evaluation workflow.
Documented uses
- Search CORD-19 paper abstracts using interactive semantic queries.
- Reuse cached corpus embeddings for later search sessions.
- Reproduce documented training results for SciBERT-NLI, BioBERT-NLI and CovidBERT-NLI.
- Fine-tune a supported transformer workflow on NLI data and evaluate it on STS.
Requirements & limits
The repository is public and specifies a GPL-2.0 license. The documented workflow runs locally and requires Python 3.6 or higher.
- Search depends on the CORD-19 data being downloaded and placed in the expected data folder.
- The first run creates corpus embeddings and may be time-consuming without a GPU; the README says both corpus and embeddings are cached for subsequent runs.
- The README identifies the application as experimental and scopes interactive browsing to paper abstracts.
- The repository’s training workflow is separate from searching: it uses NLI datasets for fine-tuning and an STS dataset for evaluation.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the local CORD-19 paper browser
Prepare the documented dependencies, model and corpus, then start the interactive semantic search demo.
Before you start
- Python 3.6 or higher
- Repository files and the libraries listed in requirements.txt
- spaCy language pack en_core_web_sm and NLTK data punkt
- An NLI-fine-tuned model downloaded from Hugging Face
- CORD-19 data downloaded from the Kaggle challenge page and placed in the data folder
Steps
Install the requirements with pip install -r requirements.txt.
Install the language resources with python -m spacy download en_core_web_sm and python -m nltk.downloader punkt.
Download a model, for example with python scripts/download_model.py --model scibert-nli.
Place the downloaded CORD-19 data in the data folder, then run python scripts/interactive_search.py.
Expected outcome
An interactive demo for querying CORD-19 paper abstracts; the corpus and embeddings are cached after the first execution.
Guide limitations
- The initial creation of embeddings for the full corpus may take time; the README suggests using a GPU.
- The documented search scope is the CORD-19 corpus and paper abstracts.
Prepare and run a documented model fine-tuning workflow
Download a pretrained model and training data, then run the repository’s NLI fine-tuning script and its documented STS evaluation.
Before you start
- Python 3.6 or higher and the repository’s installed requirements
- A pretrained model available from Hugging Face
- NLI training datasets and the STS testing dataset
Steps
Download a pretrained model, for example with python scripts/download_model.py --model scibert.
Retrieve the fine-tuning and evaluation data with python scripts/get_finetuning_data.py.
Adjust the fine-tuning parameters for the intended model, then run python scripts/finetune_nli.py.
Expected outcome
The script trains the selected model and evaluates it against the test portion of the Semantic Text Similarity benchmark dataset.
Guide limitations
- The README says parameters should be adjusted for the intended model; the default is scibert-nli.
- The documented evaluation is on the STS test portion.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- gsarti/covid-papers-browser · Official homepage title · retained in its source language