added readm

#Argumentative relation classification as plausibility ranking

come back soon

## General idea

The general idea is not confied to argumentative relation ranking, but any relation whhch can be epressed in natural language phrase.

The code is for two relation classes though extension to n classes should be possible.

In relation classification we have a SOURCE and a TARGET clause/sentence and a possible label (e.g., attack support). Let's just have an example,

``` 

SOURCE: the better a person feels , the better his brain works 

```

```

TARGET: putting physical activities in early steps of human development would finally lead to mentally healthy society

```

``` 

REL: support

```

From this, we can construct the plausible variant which signals support

```

putting physical activities in early steps of human development would finally lead to mentally healthy society. MOREOVER the better a person feels , the better his brain works

```

And the implausible variant which signals attack (in this case)

```

putting physical activities in early steps of human development would finally lead to mentally healthy society. HOWEVER the better a person feels , the better his brain works

```

## what you need

three files one for training one for dev and one for testing examples in tsv format

```

Some-ID\tabSOURCE-SENTENCE\tabTARGET-SENTENCE\tabLABEL

```

Put the file paths in data_config.py (see inside for an example)

if you want to provide context provide one file which maps the Some-ID to context of source and target:

```

Some-ID\tabSOURCE-CONTEXT\tabTARGET-CONTEXT

```

## to get results with the best model with moreover/however reconstruction and average results over five independent runs

1. adapt paths in data_config.py

2. ./run_essay.sh (if you don't have slurm, just adapt accordingly)

3. evaluate predictions: python run_eval.py

That's it.

## FAQ:

Full Aagumentaive relation Data: please shoot me an email or download from [TU Darmstadt server](https://www.informatik.tu-darmstadt.de/ukp/research_6/data/argumentation_mining_1/argument_annotated_essays_version_2/index.en.jsp) and prepare the data yourself

## Citation

```

@article{opitz2019argumentative,

  title={Argumentative Relation Classification as Plausibility Ranking},

  author={Opitz, Juri},

  journal={arXiv preprint arXiv:1909.09031},

  year={2019}

}

```