I’m wondering if there’s any fast NER models I can use that works client-side in the browser. I’m currently using a set of hard coded rules that works pretty well, but I’m wondering if I could make it work better with machine learning.
My use case is finding all entities in articles on web pages. While it’s nice if it can find the entity type, if it doesn’t correctly identify the type of some entities it’s not a big deal, the most important thing is that it finds all the entities.
When finding an entity it will start with a single word from the article, and then try to find of if that word is part of an entity and return the name of the whole entity, plus an entity type if it was able to identify it.
For example the input could be “Riku” or “Sugie” from the following sentence:
“Paidy’s president and CEO Riku Sugie and founder and executive chairman of Paidy Russel Cummer, keep their positions”
Ideally it would then return “Riku Sugie”, Person. It has to finish in less than 1 second. The entities it would need to be able to identify in this example is "Paidy, “Riku Sugie” and “Russel Cummer”.
In the sentence below it would have to be able to identify the entity “The Last Knight”. It can take more text from the article if needed, but it still has to compute the result in less than 1 second.
“For The Last Knight , critics criticized the confusing plot, messy narrative, script, characters and screen time for some characters, excessive length and constant format changes throughout the film”