Code for our TACL 2017 Submission titled "Joint Matrix-Tensor Factorization for Knowledge Base Inference" (arXiv)

Joint Matrix-Tensor Factorization for Knowledge Base Inference, Prachi Jain*, Shikhar Murty*, Mausam, Soumen Chakrabarti (*Equal Contribution)

First clone the repository:

git clone https://github.com/MurtyShikhar/KBI.git

Make sure you have Keras by running:

sudo pip install keras

Also, ensure you have the latest versions of both Theano and numpy.

To run the atomic models(where every relation is considered as an atom), for example DistMult, E or TransE, run the following:

./run.sh <dataset> <model> atomic

where MODEL can be distMult, E , complex , TransE, DATASET is wn18, fb15k, fb15k-237, nyt-fb and OPTIMIZER is either Adagrad or RMSprop

To create training data, run

dump_data.py 

and for creating data for MF, run

dump_data_pairwise.py 

with the right set of parameters. (For info on parameters, run dump_data.py -h)

For the datasets, please mail me at shikhar.murty@gmail.com.