4 ms·
Perfect timing for me. I've been working on a project that needs multilingual vectorizing and entity recognition. I've been using dbpedia and yago queries. Giv
by crucialfelix 8y ago
Perfect timing for me. I've been working on a project that needs multilingual vectorizing and entity recognition. I've been using dbpedia and yago queries.
Given an input text, what would be a good way to extract a list of entities? The word sequence should be usable to determine which is an entity or just a word.
Is there the possibility to do fine tuning a la Bert or Elmo?
- currymj 8y agoYou can use the TAGME API to do that, it’s state of the art as far as I know. If you want to implement it yourself I think it’s pretty painful, though. No reason you can’t fine tune these on your task, that’s true of any word embeddings not just Bert or Elmo.
- ikuyamada 8y agoMy past paper describes an entity linking method based on Wikipedia2Vec: https://arxiv.org/abs/1601.01343 https://arxiv.org/abs/1601.01343 You need to extract entity names using an NER software (e.g., SpaCy, Stanford NER), and resolve the names to knowledge base entities using the entity linking method.