Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/gatenlp/gateapplication-hyperpartisanclassification
GATE's Hyper-partisan scorer, based on SemEval 2019 Task 4
https://github.com/gatenlp/gateapplication-hyperpartisanclassification
Last synced: 8 days ago
JSON representation
GATE's Hyper-partisan scorer, based on SemEval 2019 Task 4
- Host: GitHub
- URL: https://github.com/gatenlp/gateapplication-hyperpartisanclassification
- Owner: GateNLP
- License: apache-2.0
- Created: 2019-08-15T10:32:51.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2021-10-27T16:21:53.000Z (about 3 years ago)
- Last Synced: 2024-11-13T18:44:05.802Z (2 months ago)
- Language: Python
- Size: 53.7 KB
- Stars: 0
- Watchers: 14
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# SemEval2019/Task4 Team Bertha-von-Suttner submission
This is the code for the [SemEval 2019 Task 4, Hyperpartisan News Detection](https://pan.webis.de/semeval19/semeval19-web/)
submitted by team `Bertha von Suttner`:
* [Ye Jiang](https://ye-jiang.github.io/)
* [Johann Petrak](http://johann-petrak.github.io)
* [Xingyi Song](http://staffwww.dcs.shef.ac.uk/people/X.Song/)
All are members of the [GATE](https://gate.ac.uk) team of the [University of Sheffield Natural Language Processing group](https://www.sheffield.ac.uk/dcs/research/groups/nlp)
The model created with this was the winning entry, see the public leaderboard (sort by accuracy column, descending):
https://www.tira.io/task/hyperpartisan-news-detection/dataset/pan19-hyperpartisan-news-detection-by-article-test-dataset-2018-12-07/
[A blog article on the GATE blog](https://gate4ugc.blogspot.com/2019/02/gate-team-wins-first-prize-in.html) briefly describes the approach taken.
If you wish to see the code as it was prepared for the SemEval 2019 task, then refer to the `semval-2019` tag in the git repo.
## Preparation / Requirements
* Python 3.6 (Anaconda will work best)
* Gensim version 3.4.0
* Tensorflow version 1.12.0
* Keras version 2.2.4
* NLTK version 3.4.1
* PyTorch version 0.4.1
* Spacy version 2.0.16
* Sklearn version 0.20.0
Once spaCy is installed, you also need to install its
[`en_core_web_sm`](https://spacy.io/usage/models) model.
Like this:
```
python -m spacy download en_core_web_sm
```
Once NLTK is installed, you also need to install its
`stopwords` data:
```
python -m nltk.downloader stopwords
```
Preparation steps:
* create a directory `elmo` and store the ELMo model files in that directory:
```
wget https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5
wget https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_options.json
```
* create a directory `data` and save the by-article training files into it:
* articles-training-byarticle-20181122.xml
* ground-truth-training-byarticle-20181122.xml
## Training
Run the following steps
* Convert the XML file into a tsv file with one article per line:
* `python -m Preprocessing.xml2line -A data/articles-training-byarticle-20181122.xml -T data/ground-truth-training-byarticle-20181122.xml -F article_sent,title_sent work/train.text.tsv`
* Convert the tsv file containing text into a tsv file containing elmo embeddings:
* If you have a GPU: `python Preprocessing/line2elmo2.py -g -l 100 work/train.text.tsv work/train.elmo.tsv`
* Otherwise: `python Preprocessing/line2elmo2.py -l 100 work/train.text.tsv work/train.elmo.tsv`
If you get problems with the GPU memory or RAM, use the -b option to reduce the batch size
* Make sure the directory `saved_models` does not contain any model files from previous runs:
* `rm saved_models/*.hdf5`
* Train the actual model:
`KERAS_BACKEND=tensorflow python CNN_elmo.py work/train.elmo.tsv`
This will create a number of model files in the `saved_models` directory. The file names contain the validation accuracy.
## Application
* Convert the XML to tsv:
`python -m Preprocessing.xml2line -A $TESTXMLFILE -F article_sent,title_sent work/test.text.tsv`
* Convert the text to elmo embeddings:
* If you have a GPU: `python Preprocessing/line2elmo2.py -g -l 100 work/test.text.tsv work/test.elmo.tsv`
* Otherwise: `python Preprocessing/line2elmo2.py -l 100 work/test.text.tsv work/test.elmo.tsv`
* Run the actual application of the model ensemble
* `./ensemble_pred.sh work/test.elmo.tsv work/test.preds.txt`
## GATE Application
The `.xgapp` file can be loaded into [GATE](https://gate.ac.uk/)
and used as a GATE application.
It requires a working [`conda`](https://conda.io/en/latest/)
environment to exist in the `condaenv` subdirectory.
The shell script `prepare-condaenv.sh` can
populate this directory from the `conda.yaml` file, and
will download the necessary NLTK data.
## ELG Service
The file `hyperpartisan_elg.py` can be used to run a web service compatible with the API specifications of the [European Language Grid (ELG)](https://european-language-grid.readthedocs.io/en/release1.1.2/all/A3_API/LTInternalAPI.html). In addition to the classifier's own dependencies, this service requires `flask` and `Flask-JSON`, and for production use you should use a production-grade WSGI server such as gunicorn rather than the built in development server of Flask.