- Harkeerat Singh Sawhney
- Yuchen Chang
- Daniela Gjorgjieva
The ability to identify the published journal for a given preprint is essential for enhancing the accessibility and citation of scientific research. Current methods are often resource-intensive or lack accuracy. Currently, Crossref achieves this mainly by using its own REST API to filter out its candidates and then using Fuzzing similarity to get predict the published Paper of a Preprint. However, this approach has its limitation if a preprint is given which is not part of the Crossref database. This report describes the Dataset creation, and the computational pipeline designed to address this issue by combining fast embeddings, clustering techniques, and LLM-based semantic similarity evaluations. We then highlight and discuss the results which we have achieved.
conda env create -f environment.ymlThere are two notebooks in the project:
dataset_creation.ipynb: This notebook contains the code for creating the dataset. It uses the Crossref API to get the published papers for the preprints in the dataset.Model Creation: This notebook contains the code for the computational pipeline. It uses the dataset created in the previous notebook to train the model and evaluate it.
- Run the
dataset_creation.ipynbnotebook to create the dataset.
- While running please make sure that you connect to the EPFL VPN or network and have a Scopus API key. The Scopus API key is required to get the abstracts of the papers from Scopus. The key should be stored in a file called
scopus_key.txtin the root directory of the project. Running the whole document you should the recreation of the dataset. The Mini Dataset is already included by default.
- Run the
Model Creationnotebook to train the model and evaluate it.
- By Running the Model Creation notebook you should be able to see the results of the model and the evaluation of the model.