CoronaWhy/team-literature-review
Functionality: Search engine developed to find articles similar to those in the target tables. Kaggle put out a table of articles curated by medical professionals (e.g. medical students extracted sample size, time periods, substrates, study population data, etc.). FAISS takes those articles and finds similar articles.
This article indicates that we can likely pair with ElasticSearch for improved results.
Goal: find similar articles to those already found in intial review (by medical experts). We need to expand this functionality to take user input and find articles similar to that input.
Input:
Output: Dataframe of metadata for articles similar to input.
Kaggle used SPECTER to create document-level embeddings. So, Christine's notebook for her FAISS similarity search engine implements those.
SPECTER: Document-level Representation Learning using Citation-informed Transformers