This project was done for a university course. The overall goal is to recommend papers to a researcher based on an initial input. The system uses BERT transformers from huggingface to create the embeddings for the data. The embeddings are created on the summaries or the preprocessed summaries. There are two different models we used for the embeddings. DistilBert and SciBert. Afterwards the similarity between the embeddings is calculated using faiss and euclidean distance.
In our experiments we wanted to test the influence of the models, preprocessing and the time it takes to find similar documents.
Results: ...
The dataset used for this can be found here