Our modular end-to-end global context-aware citation recommendation pipeline SCOPE is depicted above and entails four sub-tasks:
- Preprocessing with our Structure Analysis (see directory
PreprocessingWithStructureAnalysis) - Identification of Citation Contexts aka. cite-worthiness detection (see directory
CiteworthinesDetection) - Creating a Pool of Candidate Papers for which we are utilizing a global citation recommender (see directory
Prefetcher) - Performing the Citation Recommendation for each identified citation context (see directory
Reranker)
To get all needed dependencies, create a new environment via:
conda env create -n scope --file environment.yml
Make sure that our Modified S2ORC Dataset is accessible to the pipeline.
Run run.sh to perform the citation extraction and recommendation on all provided papers.
To change the files which are going to be loaded change the content of correct_ids.json. Alternatively you can use extraction.get_valid_ids() from the Structure Extraction module.
Notice that this selection can still contain files that can not be loaded.
- Some paper differ in the way that citation keys are written in document and the corresponding bib file
Title extraction fails if title is in format: \title[Something]{Actual Title}