UPDATE: Please see malllabiisc#69 for an issue related to a missing file.
As mentioned in Section 4.4.1 of paper, a model is trained for detecting relations associated to the question, and this model is further used in relation matching.
File pruning_main.py trains this model. A pre-trained version is available in the downloadable pretrained_models.zip
Please see the ipython notebook relation_matching_eval.ipynb for scoring QA model outputs with relation matching. This notebook assumes that the scores of the
base QA model are stored in the file webqsp_scores_full_kg.pkl. This can be done by setting writeCandidatesToFile = True in the validate function in main.py.
Then using these scores as well as the above mentioned model the final answer scores are calculated.