Is semantic similarity a good predictor of an inference relation between propositions? An attemp at training a classifier on US2016D1Reddit corpus 📖
Summary:
I calculate vector representations (meaning aproximation) of propositions from US2016D1Reddit corpus. I then calculate semantic similarity between pairs of propsitions A and B, some of which are premise-conclusion pairs while some are not. Finally I try to create a neural network which takes semantic similiarity of A and B pair as input and predicts whether A and B are premise-conlusions pairs.
For more details read the article (in Polish, unpublished) describing the project and the code:
Article_SemanticSimiliarity _and_argumentation.pdf
The code on Google Collaboratory:
https://colab.research.google.com/drive/13lJb_aUPstlC6i_F1e-xcNk8S9H4Dw_f?usp=sharing
The main idea and parts of code are taken from:
https://github.com/atka555/argument-mining
In order to run the code yourself, you need to download the "corpus_US2016D1reddit.zip" file. Next, in project code you need to change the line "maps = glob.glob("/US2016D1reddit/\*.json")" to match the location of the file on your computer.
If you don't want to use Google Collaboratory you can donwload the .py file provided in this repository.
