Haystack docs home page

Utilizing existing FAQs for Question Answering


While extractive Question Answering works on pure texts and is therefore more generalizable, there's also a common alternative that utilizes existing FAQ data.


  • Very fast at inference time
  • Utilize existing FAQ data
  • Quite good control over answers


  • Generalizability: We can only answer questions that are similar to existing ones in FAQ

In some use cases, a combination of extractive QA and FAQ-style can also be an interesting option.

# Install the latest release of Haystack in your own environment 
#! pip install farm-haystack

# Install the latest master of Haystack and install the version of torch that works with the colab GPUs
!pip install git+https://github.com/deepset-ai/haystack.git
!pip install torch==1.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html
from haystack import Finder
from haystack.document_store.elasticsearch import ElasticsearchDocumentStore

from haystack.retriever.dense import EmbeddingRetriever
from haystack.utils import print_answers
import pandas as pd
import requests

Start an Elasticsearch server

You can start Elasticsearch on your local machine instance using Docker. If Docker is not readily available in your environment (eg., in Colab notebooks), then you can manually download and execute Elasticsearch from source.

# Recommended: Start Elasticsearch using Docker
# ! docker run -d -p 9200:9200 -e "discovery.type=single-node" elasticsearch:7.6.2
# In Colab / No Docker environments: Start Elasticsearch from source
! wget https://artifacts.elastic.co/downloads/elasticsearch/elasticsearch-7.6.2-linux-x86_64.tar.gz -q
! tar -xzf elasticsearch-7.6.2-linux-x86_64.tar.gz
! chown -R daemon:daemon elasticsearch-7.6.2

import os
from subprocess import Popen, PIPE, STDOUT
es_server = Popen(['elasticsearch-7.6.2/bin/elasticsearch'],
                   stdout=PIPE, stderr=STDOUT,
                   preexec_fn=lambda: os.setuid(1)  # as daemon
# wait until ES has started
! sleep 30

Init the DocumentStore

In contrast to Tutorial 1 (extractive QA), we:

  • specify the name of our text_field in Elasticsearch that we want to return as an answer
  • specify the name of our embedding_field in Elasticsearch where we'll store the embedding of our question and that is used later for calculating our similarity to the incoming user question
  • set excluded_meta_data=["question_emb"] so that we don't return the huge embedding vectors in our search results
from haystack.document_store.elasticsearch import ElasticsearchDocumentStore
document_store = ElasticsearchDocumentStore(host="localhost", username="", password="",

Create a Retriever using embeddings

Instead of retrieving via Elasticsearch's plain BM25, we want to use vector similarity of the questions (user question vs. FAQ ones). We can use the EmbeddingRetriever for this purpose and specify a model that we use for the embeddings.

retriever = EmbeddingRetriever(document_store=document_store, embedding_model="deepset/sentence_bert", use_gpu=False)

Prepare & Index FAQ data

We create a pandas dataframe containing some FAQ data (i.e curated pairs of question + answer) and index those in elasticsearch. Here: We download some question-answer pairs related to COVID-19

# Download
temp = requests.get("https://raw.githubusercontent.com/deepset-ai/COVID-QA/master/data/faqs/faq_covidbert.csv")
open('small_faq_covid.csv', 'wb').write(temp.content)

# Get dataframe with columns "question", "answer" and some custom metadata
df = pd.read_csv("small_faq_covid.csv")
# Minimal cleaning
df.fillna(value="", inplace=True)
df["question"] = df["question"].apply(lambda x: x.strip())

# Get embeddings for our questions from the FAQs
questions = list(df["question"].values)
df["question_emb"] = retriever.embed_queries(texts=questions)
df["question_emb"] = df["question_emb"].apply(list) # convert from numpy to list for ES ingestion
df = df.rename(columns={"answer": "text"})

# Convert Dataframe to list of dicts and index them in our DocumentStore
docs_to_index = df.to_dict(orient="records")

Ask questions

Initialize a Finder (this time without a reader) and ask questions

finder = Finder(reader=None, retriever=retriever)
prediction = finder.get_answers_via_similar_questions(question="How is the virus spreading?", top_k_retriever=10)
print_answers(prediction, details="all")