• The path ranking algorithm (PRA) and subgraph feature extraction (SFE) are algorithms that extract feature matrices from graphs. This git repository contains code implementating these two algorithms, produces matrices that are useful in a variety of situations. The repository also contains code for running experiments doing link prediction in a graph using these feature matrices (also known as knowledge base completion, when the graph corresponds to a knowledge base such as Freebase or NELL).

  • Using feature matrices generated from graphs (using the PRA/SFE code linked above) to answer questions, both science questions and machine-generated questions from the Facebook bAbI dataset.