Text Summarization models

if you are able to endorse me on Arxiv, i would be more than glad https://arxiv.org/auth/endorse?x=FRBB89 thanks This repo is built to collect multiple implementations for abstractive approaches to address text summarization , for different languages (Hindi, Amharic, English, and soon isA Arabic)

If you found this project helpful please consider citing our work, it would truly mean so much for me

@inproceedings{Deep-Architectures-Abstractive-Text-Summarization-Amr-Zaki-2019,
	title = "Deep Architectures for Abstractive Text
	Summarization in Multiple Languages",
	author = "Amr M. Zaki  and
	Mahmoud I. Khalil  and
	Hazem M. Abbas.",
	booktitle = "14th IEEE International Conference on Computer Engineering and Systems (ICCES 2019)",
	month = dec,
	year = "2019",
	address = "Cairo, Egypt",
	publisher = "IEEE",
	url = "http://www.icces.org.eg/",
	doi = "",
	pages = "",
}

it is built to simply run on google colab , in one notebook so you would only need an internet connection to run these examples without the need to have a powerful machine , so all the code examples would be in a jupiter format , and you don't have to download data to your device as we connect these jupiter notebooks to google drive

• Arabic Summarization Model using the corner stone implemtnation (seq2seq using Bidirecional LSTM Encoder and attention in the decoder) for summarizing Arabic news
• implementation A Corner stone seq2seq with attention (using bidirectional ltsm ) , three different models for this implemntation
• implementation B seq2seq with pointer genrator model
• implementation C seq2seq with reinforcement learning

Blogs

This repo has been explained in a series of Blogs

• to understand how to work with google colab eco system , and how to integrate it with your google drive , this blog can prove useful DeepLearning Free Ecosystem
• Tutorial 1 Overview on the different appraches used for abstractive text summarization
• Tutorial 2 How to represent text for our text summarization task
• Tutorial 3 What seq2seq and why do we use it in text summarization
• Tutorial 4 Multilayer Bidirectional Lstm/Gru for text summarization
• Tutorial 5 Beam Search & Attention for text summarization
• Tutorial 6 Build an Abstractive Text Summarizer in 94 Lines of Tensorflow
• Tutorial 7 Pointer generator for combination of Abstractive & Extractive methods for Text Summarization
• Tutorial 8 Teach seq2seq models to learn from their mistakes using deep curriculum learning
• Tutorial 9 Deep Reinforcement Learning (DeepRL) for Abstractive Text Summarization made easy
• Tutorial 10 Hindi Text Summarization

---

Try out this text summarization through this website (eazymind) , which enables you to summarize your text through

• curl call

curl -X POST 
http://eazymind.herokuapp.com/arabic_sum/eazysum
-H 'cache-control: no-cache' 
-H 'content-type: application/x-www-form-urlencoded' 
-d "eazykey={eazymind api key}&sentence={your sentence to be summarized}"

• python package (pip install eazymind) pip install eazymind

from eazymind.nlp.eazysum import Summarizer

#---key from eazymind website---
key = "xxxxxxxxxxxxxxxxxxxxx"

#---sentence to be summarized---
sentence = """(CNN)The White House has instructed former
    White House Counsel Don McGahn not to comply with a subpoena
    for documents from House Judiciary Chairman Jerry Nadler, 
    teeing up the latest in a series of escalating oversight 
    showdowns between the Trump administration and congressional Democrats."""
    
summarizer = Summarizer(key)
print(summarizer.run(sentence))

---

Implementation A (seq2seq with attention and feature rich representation)

contains 3 different models that implements the concept of hving a seq2seq network with attention also adding concepts like having a feature rich word representation This work is a continuation of these amazing repos

Model 1

is a modification on of David Currie's https://github.com/Currie32/Text-Summarization-with-Amazon-Reviews seq2seq

Model 2

1- Model_2/Model_2.ipynb

a modification to https://github.com/dongjun-Lee/text-summarization-tensorflow

2- Model_2/Model 2 features(tf-idf , pos tags).ipynb

a modification to Model 2.ipynb by using concepts from http://www.aclweb.org/anthology/K16-1028

Results

A folder contains the results of both the 2 models , from validation text samples in a zaksum format , which is combining all of

• bleu
• rouge_1
• rouge_2
• rouge_L
• rouge_be for each sentence , and average of all of them

Model 3

a modification to https://github.com/thomasschmied/Text_Summarization_with_Tensorflow/blob/master/summarizer_amazon_reviews.ipynb

---

Implementation B (Pointer Generator seq2seq network)

it is a continuation of the amazing work of https://github.com/abisee/pointer-generator https://arxiv.org/abs/1704.04368 this implementation uses the concept of having a pointer generator network to diminish some problems that appears with the normal seq2seq network

Model_4_generator_.ipynb

uses a pointer generator with seq2seq with attention it is built using python2.7

zaksum_eval.ipynb

built by python3 for evaluation

Results/Pointer Generator

• output from generator (article / reference / summary) used as input to the zaksum_eval.ipynb
• result from zaksum_eval

i will still work on their implementation of coverage mechanism , so much work is yet to come if God wills it isA

---

Implementation C (Reinforcement Learning For Sequence to Sequence )

this implementation is a continuation of the amazing work done by https://github.com/yaserkl/RLSeq2Seq https://arxiv.org/abs/1805.09461

@article{keneshloo2018deep,
 title={Deep Reinforcement Learning For Sequence to Sequence Models},
 author={Keneshloo, Yaser and Shi, Tian and Ramakrishnan, Naren and Reddy, Chandan K.},
 journal={arXiv preprint arXiv:1805.09461},
 year={2018}
}

Model 5 RL

this is a library for building multiple approaches using Reinforcement Learning with seq2seq , i have gathered their code to run in a jupiter notebook , and to access google drive built for python 2.7

zaksum_eval.ipynb

built by python3 for evaluation

Results/Reinforcement Learning

• output from Model 5 RL used as input to the zaksum_eval.ipynb