DensePhrases is an open domain Q&A technology created by Jinhyuk Lee at Korea University, and was published as a paper titled “Learning Dense Representations of Phrases at Scale”. Here is a link to the paper:
problem, without the need for processing documents on-demand during inference
(Seo et al., 2019). However, current phrase retrieval models heavily depend on
their sparse representations while still underperforming retriever-rea…
When a question is given, it selects the most suitable paragraph from about 60 billion Wikipedia paragraphs and extracts key words suitable for the question. It has the characteristic of searching only with a delay of about 100ms using only one server. Here is a link to the Github repository:
In particular, it features an effective indexing technique for quick search from large data, and includes a link to a demo site where you can test it yourself online. It is not possible to test it in Korean because it is learned about English data, but the Github repository contains links to thesis as well as data and codes, so it will be very helpful for related research. Here is a link to DensePhrases' online test page:
