Question 1

How do I set up DrQA with the CoreNLP tokenizer?

Accepted Answer

Download Stanford CoreNLP jars, set the CLASSPATH environment variable, and run the install_corenlp.sh script. Verify with the Python code snippet in the README to ensure the tokenizer loads correctly.

Question 2

DrQA vs BERT for question answering: which is better?

Accepted Answer

DrQA is a full pipeline with retrieval and reading, ideal for open-domain QA over large corpora, while BERT excels at comprehension tasks. DrQA's strength is scalability, but BERT-based models may offer higher accuracy on pure reading benchmarks.

Question 3

Can DrQA handle questions about current events?

Accepted Answer

No, because it uses a static Wikipedia dump from 2016. To handle recent events, you'd need to update the document database and retrain the models, which requires significant effort and computational resources.

Question 4

How to train DrQA on my own documents?

Accepted Answer

Preprocess documents into the SQLite database format, build a new TF-IDF retriever using the retriever scripts, and fine-tune the reader on your data with distant supervision or SQuAD-style training, as described in the component READMEs.

Question 5

What are DrQA's performance benchmarks on SQuAD?

Accepted Answer

The single model achieves 69.4 EM and 78.9 F1 on SQuAD dev, while the multitask model scores 29.5 EM in the full Wikipedia setting, as detailed in the trained models section with specific evaluation metrics.

Question 6

Is DrQA suitable for commercial applications?

Accepted Answer

It can serve as a baseline for QA systems, but consider its limitations: static retrieval, dependency on large resources, and extractive-only answers, which may not meet all production needs without customization.

DrQA

What is DrQA?

Overview

Use Cases

Best For

Related Projects

Found a gem we're missing?

Not Ideal For

Pros & Cons

Pros

Cons

Frequently Asked Questions