Closed dg4271 closed 5 years ago
The subgraph contains all facts that are within two steps away from the question entities. If there are too many facts, we compute their personalized pagerank score, and pick the top few ones.
So, Gregoire Colin appears in Before the Rain, which is of distance one. And facts with Before the Rain as subject have distance of two. Thats why they are included.
We would like to build a high recall system. So it's fine to include unrelated facts in the subgraph.
We are planning to clean up the preprocessing code and release it. Please let me know if you have more questions.
Thank a lot for your excellent work! Just want to try this model out on some other open-source corpus, so the preprocessing code would be a great help.
Could you please let us know are you planning to release the preprocessing code in the near future? Thanks A LOT! 😜
Thank you for sharing your great work!
But i want to know more detail about preprocessing. (e.g. how to make sub-graph from a question sentence) Because, i wonder why a tuple <Before the Rain, starred_actors, Katrin Cartlidge> is in sub-graph of a question "what does Grégoire Colin appear in ?"