yerkesoul / subjectivity-classification

Replication and extention of Alhindi et al. (2020) for course "Mining opinions and arguments".

Subjectivity-Classification.-Identifying-news-and-editorials

Replication and extention of for Alhindi et al. (2020) for course "Mining opinions and arguments".

The replication and further extension of argumentation features model.

The main goal is to investigate the importance of article content representation with argumentation features in distinguishing between news and opinion articles.The results of the original model and replication differ as two models were not trained on the same model. However, argumentation features were improving the result of the model almost always. Moreover, the frequency analysis of argumentation features confirms the relation between argumentative types of sentences and the discourse structure of the articles. The report describes details of the project.

The Models folder contains the notebooks:

1.Arg_Features_Bert_Embeddings.ipynb contains models such as:

• BERT,
• RNN (3 labels)
• RNN (6 labels)
• BERT+RNN (3 labels)
• BERT+RNN (6 labels)

2.Vader_POS_merged.ipynb contains models such as:

• VADER
• POS tagger
• VADER+POS tagger

3.All_4_models_combined.ipynb contains models such as:

• BERT+RNN+ VADER+POS tagger (3 labels)
• BERT+RNN+VADER+POS tagger (6 labels)

The Extracting features contains scripts needed for extracting BERT embeddings and argumentation features:

1.ExtractingBertEmbeddings.ipynb derives BERT embeddings with [CLS] token representing the whole article.The embeddings are saved separately per ariticle and collected in a folder, further the folder is wrapped in a zip format.

2.TraingArgClassification.ipynb file is used to train the BERT argumentation classification model and the weights are saved for the later extraction.

3.ExtractingArgFeatures.ipynb is needed for prediction and extraction of arg. features.

Data has the file for the dataset preprocessing, however due to data protection law they can not be shown publickly.

Lastly,there is a Plots.ipynb which shows the realation of argumentative types of sentences and the discourse structure of the articles.