Take a seed set of topics and arxiv paper links. Get related papers, pass them into pretrained LM Plot in reduced dimension Propagate and obtain more relevant papers and strive to get a better recommender system for papers. Modular options [paper fetch] : co-authors, same domain, by time, custom [embeddings] : BERT, GPT, custom-model [dim-reduction] : PCA, t-SNE, Embedding layer [similarity-measure] : cosine, jaccard, euclidean, manhattan