Awesome Open Source
Awesome Open Source


PyPI version

Giveme5W1H is an open source, state of the art system that extracts phrases answering the journalistic 5W1H questions describing a news article's main event, i.e., who did what, when, where, why, and how? You can access the system through a simple RESTful API from any programming language or use it as a Python 3 library.

The figure below shows an excerpt of a news article with highlighted 5W1H phrases.

Getting started

It's super easy, we promise!


Giveme5W1H requires Python 3.6 (or later) to run. The following two commands will install Giveme5W1H and Stanford CoreNLP Server.

$ pip3 install giveme5w1h # Linux users may need to use: sudo -H pip3 install giveme5w1h
$ giveme5w1h-corenlp install

Test if the Stanford CoreNLP Server setup was successful:

$ giveme5w1h-corenlp

You should see [main] INFO CoreNLP - StanfordCoreNLPServer listening at /0:0:0:0:0:0:0:0:9000 after a couple of seconds. Exit the program by pressing CTRL+C.

Extract 5W1H Phrases

Giveme5W1H enables the extraction of 5W1H phrases from news articles. You can access Giveme5W1H's functionality via a RESTful API, or as a module from within your Python 3.6+ code.

Starting the CoreNLP Server (mandatory)

You must always start the Stanford CoreNLP Server before using Giveme5W1H. To do so, run the following command in a terminal, and do not close the terminal:

$ giveme5w1h-corenlp

Use within your own code (as a library)

Use the following code to extract 5W1H phrases from a single news article.

from Giveme5W1H.extractor.document import Document
from Giveme5W1H.extractor.extractor import MasterExtractor

extractor = MasterExtractor()
doc = Document.from_text(text, date_publish)
# or: doc = Document(title, lead, text, date_publish) 
doc = extractor.parse(doc)

top_who_answer = doc.get_top_answer('who').get_parts_as_text()

Have a look at our sample Python scripts, for more information on extraction from a single news article, or a folder consisting of multiple JSON files in news-please format. To start one of them, use the following command (here shown for the parse_documents script, which extracts 5W1Hs from multiple JSON files):

python3 -m Giveme5W1H.examples.extracting.parse_documents

RESTful API / webpage access

Start the RESTful API server that comes with Giveme5W1H (execute the following command in a separate shell, so that the CoreNLP Server started by the previous command runs in parallel):

$ giveme5w1h-rest

After a couple of seconds, you will see the following line:

 * Running on (Press CTRL+C to quit)

If you open the URL in your browser, you will see a page with a sample news article. Just click on GET example, or run example to analyze the shown article. You can also use this page to analyze your articles.

Of course, you can also access the RESTful API endpoints directly. You can access the endpoint at http://localhost:9099/extract via GET or POST requests. For GET and POST requests, the input format is the news-please article format, with the following fields:

  • title (mandatory; can also be used to pass the full text of the article, e.g., if you do not have title, description, and text separately)
  • description (typically the lead paragraph)
  • text (the main text)
  • date (must be readable by parsedatetime, e.g., 2017-07-17 17:03:00)
  • url (mandatory for POST requests)

Additional Information

This section is currently subject to a major update. Some information may be outdated or redundant to the above information.

Why do I need to start the Stanford CoreNLP Server manually?

We decided to not integrate the CoreNLP Server transparently into Giveme5W1H mainly because the CoreNLP Server takes a lot of time until the initialization of all components is finished. Hence, the first run of Giveme5W1H after you started the CoreNLP Server, will likely take a couple of minutes (because components in CoreNLP Server are initialized on the fly). So, be sure to start the server and use it to extract 5W1Hs from multiple news articles, or - even better - have the CoreNLP Server run permanently. See below if you want to use a CoreNLP Server that is running on a remote machine or different port.


The following configurations are optional.

CoreNLP Host

You can also use a remotely installed CoreNLP-Server. Simply parse the preprocessor another URL in case you run it on another machine:

from Giveme5W1H.extractor.preprocessors.preprocessor_core_nlp import Preprocessor

preprocessor = Preprocessor('')


  • For file-based data - every input is transferred to the output
    • For instance, annotated is already a part of the provided example files
  • Each Question has their extracted candidates under extracted,
  • Each Candidate, has parts, score and text property and their sentence index.
  • Each part is structured as (payload, Postoken)
  • Each payload has at least nlpToken which is the "basic" information.
  • Each enhancer is saving his information under their name in the payload

See the sample.json for details.


Giveme5W can read and write only in a JSON format example. You find ready to use examples here

dID is used for matching input and output, not the filename!

There is an easy-to-use handler to work with files; these are all options::

 documents = (
        # initiate the file handler with the input directory
            # add giveme5w extractor  (it would only copying files without...)

            # Optional: set a output directory

            # Optional: set a path to cache and load preprocessed documents (CoreNLP & Enhancer results)

            # Optional: limit the documents read from the input directory (handy for development)

            # Optional: resume ability, skip input file if its already in output

            # load and saves all document runtime objects for further programming

            ## setup is done: executing it

            # get the processed documents, this can only be done because preload_and_cache_documents was called

Check the examples under and for more details


CoreNLP and Enhancer have a long execution time; therefore it is possible to cache the result at the filesystem to speed up multiple executions. Delete all files in "/cache", if you want to process them again, see examples in 'examples/extracting' for more details.

if you add or remove enhancer, you must delete all files in the cache directory (if caching is enabled (set_preprocessed_path))


/examples/misc/ is running the extractor with different weights from 0-10. The best candidate is compared with the best annotation to get a score. The calculated score, document id, and the used weights are saved per question under ./results.

Because of the combined_scorer, each document is evaluated in each step. This can lead to entries with the same weights, but with different scores.

How to cite

If you are using Giveme5W1H, please cite our paper:

  author    = {Hamborg, Felix and Breitinger, Corinna and Gipp, Bela},
  title     = {Giveme5W1H: A Universal System for Extracting Main Events from News Articles},
  booktitle = {Proceedings of the 13th ACM Conference on Recommender Systems, 7th International Workshop on News Recommendation and Analytics (INRA 2019)},
  year      = {2019},
  month     = {Sept.},
  location  = {Copenhagen, Denmark}

You can find more information on this and other news projects on our website.

Contribution and support

Do you want to contribute? Great, we are always happy for any support on this project! Just send a pull request. By contributing to this project, you agree that your contributions will be licensed under the project's license (see below). If you have questions or issues while working on the code, e.g., when implementing a new feature that you would like to have added to Giveme5W1H, open an issue on GitHub and we'll be happy to help you. Please note that we usually do not have enough resources to implement features requested by users - instead we recommend to implement them yourself, and send a pull request. Please understand that we are not able to provide individual support via email. We think that help is more valuable if it is shared publicly so that more people can benefit from it.


Licensed under the Apache License, Version 2.0 (the "License"); you may not use Giveme5W1H except in compliance with the License. A copy of the License is included in the project, see the file LICENSE.txt.

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

The Giveme5W1H logo is courtesy of Mario Hamborg.

Copyright 2018-2020 The Giveme5W1H team

Related Awesome Lists
Top Programming Languages
Top Projects

Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
Html (247,396
Server (66,089
Rest (29,263
Natural Language Processing (14,982
Article (13,945
Stanford (2,307
Text Analysis (870
Questions And Answers (748
News Articles (601
Nlp Library (208
Questions (180
Event Detection (45
Event Extraction (44
Answering (3