【转】NLP for Question Answering: IR-QA
Over the course of the next two months, two of Cloudera Fast Forward’s Research Engineers,
Melanie Beck and Ryan Micallef, will build a QA system following the information retrieval-based
method, by creating a document retriever and document reader.
We’ll focus our efforts on exploring and experimenting with various Transformer architectures
(like BERT) for the document reader, as well as off-the-shelf search engine algorithms for the retriever.
IR-QA
-
-Below we illustrate the workflow of a generic IR-based QA system. These systems generally
have two main components: the document retriever and the document reader.
The document retriever functions as the search engine, ranking and retrieving relevant
documents to which it has access. It supplies a set of candidate documents that could
answer the question (often with mixed results, per the Google search shown above).
The document reader consists of reading comprehension algorithms built with core
NLP techniques. This component processes the candidate documents and extracts
from one of them an explicit span of text that best satisfies the query. Let’s dive
deeper into each of these components.
======================
Ref:
https://experiments.fastforwardlabs.com/
https://qa.fastforwardlabs.com/
Building a QA System with BERT on Wikipedia
https://qa.fastforwardlabs.com/pytorch/hugging%20face/wikipedia/bert/transformers/2020/05/19/Getting_Started_with_QA.html