Neural network Connected Papers

Connected Papers is a unique visual tool that helps researchers and practitioners find and explore documents relevant to their field.

• To create each graph, CP analyzes about ~50,000 articles and selects the few dozen that have the strongest connections to the original article.
• In the graph, articles are arranged according to their similarity . This means that even articles that do not directly cite each other can be closely related and very close together. Related articles are not a citation tree.
• The similarity metric is based on the concepts of co-citation and bibliographic linkage . This metric assumes that two articles with highly overlapping citations and references are more likely to cover a related topic.
• The algorithm then builds a force-directed graph to distribute articles in a way that visually groups similar articles together and distances less similar ones from each other. When selecting a node, we select the shortest path from each node to the source article in the similarity space.
• The database is connected to the Semantic Scholar document corpus (licensed under ODC-BY). Their team has done an amazing job of compiling hundreds of millions of published articles in many scientific fields.