https://github.com/udibr/headlines

Automatically generate headlines to short articles
https://github.com/udibr/headlines

generation keras nlp rnn summarization

Last synced: 6 months ago
JSON representation

Automatically generate headlines to short articles

• Host: GitHub
• URL: https://github.com/udibr/headlines
• Owner: udibr
• License: mit
• Created: 2016-04-21T02:23:53.000Z (over 9 years ago)
• Default Branch: master
• Last Pushed: 2018-04-22T17:32:41.000Z (over 7 years ago)
• Last Synced: 2024-08-08T23:20:28.824Z (over 1 year ago)
• Topics: generation, keras, nlp, rnn, summarization
• Language: Jupyter Notebook
• Size: 634 KB
• Stars: 526
• Watchers: 28
• Forks: 150
• Open Issues: 30
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Automatically generate headlines to short articles



This project attempts to reproduce the results in the paper:

[Generating News Headlines with Recurrent Neural Networks](http://arxiv.org/abs/1512.01712)

## How to run

### Software

* The code is running with [jupyter notebook](http://jupyter.org/)

* Install [Keras](http://keras.io/)

* `pip install python-Levenshtein`

### Data

It is assumed that you already have training and test data.

The data is made from many examples (I'm using 684K examples),

each example is made from the text

from the start of the article, which I call description (or `desc`),

and the text of the original headline (or `head`).

The texts should be already tokenized and the tokens separated by spaces.

Once you have the data ready save it in a python pickle file as a tuple:

`(heads, descs, keywords)` were `heads` is a list of all the head strings,

`descs` is a list of all the article strings in the same order and length as `heads`.

I ignore the `keywords` information so you can place `None`.

### Build a vocabulary of words

The [vocabulary-embedding](./vocabulary-embedding.ipynb)

notebook describes how a dictionary is built for the tokens and how

an initial embedding matrix is built from [GloVe](http://nlp.stanford.edu/projects/glove/)

### Train a model

[train](./train.ipynb) notebook describes how a model is trained on the data using [Keras](http://keras.io/)

### Use model to generate new headlines

[predict](./predict.ipynb) generate headlines by the trained model and

showes the attention weights used to pick words from the description.

The text generation includes a feature which was

not described in the original paper, it allows for words that are outside

the training vocabulary to be copied from the description to the generated headline.

## Examples of headlines generated

Good (cherry picking) examples of headlines generated

![cherry picking of generated headlines](./cherry_picking.png)

![cherry picking of generated headlines](./cherry_picking1.png)

## Examples of attention weights

![attention weights](./attention_weights.png)