After reading the KB Retrieval part of your paper, I got a question. You mention "run the Personalized PageRank around these seeds..". You also mention that using semantics similarity to give weight to different vectors.
Can I paraphrase that you run a edge-weighted Personalized PageRank, with weights based on semantic similarity between relation and question?
Yes, that is correct. The weights were obtained using cosine similarity between the average word embeddings for the question and average word embeddings for the surface form of the relation.
Hi,
After reading the KB Retrieval part of your paper, I got a question. You mention "run the Personalized PageRank around these seeds..". You also mention that using semantics similarity to give weight to different vectors.
Can I paraphrase that you run a edge-weighted Personalized PageRank, with weights based on semantic similarity between relation and question?
Thank you very much!