Implementation of the automatic alignment model from the paper "Aligning Actions Across Recipe Graphs". In this paper we present two automatic alignment models (base, extended) and a simple baseline (cosine similarity).

You can find all the requirements in the file requirement.txt.

• Python 3.7
• Conllu 4.2.2
• Matplotlib 3.3.2
• Pandas 1.2.3
• Pytorch 1.7.1
• Transformers 4.0.0
• Flair 0.8.0.post1
• AllenNLP 0.9.0

The Alignment Model requires recipes parsed into action graphs in CoNLL-U format and alignment information for action pairs presented in tsv format (see ../data).

For tagging raw recipe text and parsing tagged recipes into dependency graphs, we used the models described in ./preprocessing.

Download the corpus from here into ./data folder for reproducing our experiment results. Additionally, create the results folder where the trained models and their test results will be saved (Notes: You can change the hyperparameters and the path names in the file constants.py). Per default, the script looks for the following results folders:

To reproduce our experimental results, run the following command from this directory:

python main.py [model_name] --embedding_name [embedding_name]

where [model_name] could be one of the following:

• Sequence : Sequential Ordering of Alignments
• Simple : Cosine model (Baseline)
• Naive : Common Action Pair Heuristics mode (Naive Model)
• Alignment-no-feature : Base Alignment model (w/o parent+child nodes)
• Alignment-with-feature : Extended Alignment model (with parent+child nodes)

and [embedding_name] could be one of the following:

• bert : BERT embeddings (default)
• elmo : ELMO embeddings

Our experiment results are as follows:

Both the base and the extended Alignment models were trained for 10 folds each with 40 epochs.