Rynchops

A Fast Abstractive Summarization & Translation Demo App

Rynchops niger (Black Skimmer) skimming in flight by Charles J. Sharp. Skimmers fly low and fast aross the surface of the water with their lower beak submerged to catch and eat fish. Image from Wikimedia Commons, available under CC BY-SA 4.0.

This repository contains an app to demonstrate the proof-of-concept for the Fast Abstractive Summarization & Translation project. The app is a streamlit application that runs in the browser. It utilizes a number of machine learning models.

Running The Application

Prerequisites

The Rynchops app requires several python packages. To run the app, one needs Python and a package manager. We use Conda as our package manager.

• Python 3.7+
• Conda Package Manager

Obtaining This Code

To run the app, you will need this code. The best way to obtain the code is to "clone" it. This is a proccess that copies the code to your computer and sets up a git repository in it. This allows you to get updates to the code whenever this repository is updated, and allows you to contribute your own code to this repository. To clone a git repository, please refer to the official documentation, "Cloning a repository".

Creating the Conda Environment

The packages for the app are listed in environment.yml. You will need to install those packages with conda. To do so, once you have installed conda and python and downloaded the , you may create an environment containing the packages with the following command.

conda create -f environment.yml

POCs

The points of contact for this project are Elias Jaffe & Mike Brasseur.

(c) ATS 2021

Prod

conda activate rynchops
streamlit run app.py

Roadmap

Phase 1: Demo

Input -> Chunker for variable size output -> Same-language Summarizer Model -> Translator -> Output
                              |-> NER 

1. Model Pipelining
2. Demo Application Development

Phase 2: Product Development

Input -> Chunker for variable size output -> Cross-language Summarizer Model -> Output
                               | -> NER

1. Data Collection
2. Dataset labeling
3. Model training
4. Develop Service Platform