Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/dataqa/nlp-labelling

Labelling platform for text using weak supervision.
https://github.com/dataqa/nlp-labelling

annotation-tool data-labeling data-science learning-with-limited-labeled-data learning-with-noisy-labels natural-language-processing ner nlp nlp-machine-learning pseudo-labeling search-engine text-annotation-tool text-classification text-mining weak-supervision

Last synced: 14 days ago
JSON representation

Labelling platform for text using weak supervision.

Awesome Lists containing this project

README 

        


    

    
DataQA







    

    

    

    




 



DataQA is a tool to label and explore unstructured documents. It uses rules-based weak supervision to significantly reduce the number of labels needed compared to other tools. Here are a few things you can do with it:

- Search your documents using Elasticsearch powerful text search engine,

- Classify your documents,

- Extract entities from your own data or from Wikipedia,

- Link mentions of entities to your own ontology.



... and it's all available with a simple pip command!



 





    




 



* [Installation](#installation)

* [Usage](#usage)

* [What is weak supervision and why does it work?](#what-is-weak-supervision-and-why-does-it-work)

* [Tutorials](#documentation)

* [Contact](#contact)



# Installation



## Pre-requisites:



* Python 3.6, 3.7, 3.8 and 3.9

* (Recommended) start a new python virtual environment

* Update your pip `pip install -U pip`

* Tested on backend: MacOSX, Ubuntu. Tested on browser: Chrome, Firefox.



## Installing from pypi



* `pip install dataqa`



## To run with Docker



* The first time it is run: `docker run -d -p 5000:5000 dataqa/dataqa`

* In order to keep the data between runs, use `docker start [container-id]` and `docker stop [container-id]`



# Usage



In the terminal, type `dataqa run`. Wait a few minutes initially, as it takes some minutes to start everything up.



Doing this will run a server locally and open a browser window at port `5000`. If the application does not open the browser automatically, open `localhost:5000` in your browser. You need to keep the terminal open.



To quit the application, simply do `Ctr-C` in the terminal. To resume the application, type `dataqa run`. Doing so will create a folder at `$HOME/.dataqa_data`.



## Uploading data



The text file needs to be a csv file in utf-8 encoding of up to 30MB with a column named "text" which contains the main text. The other columns will be ignored.



This step is running some analysis on your text and might take up to 5 minutes.



## Uninstall



In the terminal:



* `dataqa uninstall`: this deletes your local application data in the home directory in the folder `.dataqa_data`. It will prompt the user before deleting.

* `pip uninstall dataqa` 



### Does this tool need an internet connection?



Nope. **No data will ever leave your local machine.**



## Troubleshooting



If the project data does not load, try to go to the homepage and `http://localhost:5000` and navigate to the project from there.



Try running `dataqa test` to get more information about the error, and bug reports are very welcome!



To test the application, it is possible to upload a text that contains a column "\_\_LABEL\_\_". The ground-truth labels will then be displayed during labelling and the real performance will be shown in the performance table between brackets.



# Documentation



Documentation at: [https://dataqa.ai/docs/](https://dataqa.ai/docs/).



* To get started with a multi-class classification problem, go [here](https://dataqa.ai/docs/latest/tutorials/ecomm_product_categories/classification_product_categories/).

* To get started with a named entity recognition problem, go [here](https://dataqa.ai/docs/latest/tutorials/medical_side_effects/ner_medical/).

* To get started with a named entity linking problem, go [here](https://dataqa.ai/docs/latest/tutorials/medical_entity_disambiguation/ned_side_effects/).



# What is weak supervision and why does it work?



Weak supervision is a set of techniques to produce noisy labels for large quantities of data. It has gained popularity in recent years due to the large amounts of data typically needed for ML systems. The annotator is able to encode any prior domain knowledge it has in the form of rules. Even though these rules can be noisy, the algorithm learns how to weigh them accordingly and use them as signals to extract patterns from the data.





    
Creating a rule for classification


    

    
 


    
Creating a rule for NER


    




# Contact



For any feedback, please contact us at [email protected]. Also follow me on [![alt text][1.1]][1] for more updates and content around ML and labelling.



[1.1]: https://i.imgur.com/wWzX9uB.png 

[1]: https://www.twitter.com/DataqaAi