Accompanying code for Violence Rating Prediction from Movie Scripts

Folder Structure

• experiments: Contains the code to run the experiments
• notebooks: Model error analysis
• lexicons: Contains the folders to store the lexicons

Instructions:

Download Lexicons:

1. AFINN-111 into lexicons/AFINN from [AFINN](http://www2.imm.dtu.dk/pubdb/views/publication_details.php?id=6010)
2. categories.tsv into lexicons/empath from [Empath](https://github.com/Ejhfast/empath-client)
3. hatebase_dict.csv and refined_ngram_dict.csv into lexicons/hatespeech_davidson from <https://github.com/t-davidson/hate-speech-and-offensive-language>
4. expandedLexicon.txt into lexicons/lexicon_abusive_words from <https://github.com/uds-lsv/lexicon-of-abusive-words>
5. vader_lexicon.txt into lexicon/vader from <https://github.com/cjhutto/vaderSentiment>

Experiments

The experiments folder contains the scripts to replicate the experiments presented in the paper. All models were trained on cross-validated fashion with the folds pre-calculated and stored in hard drive. We added a bash script to ease the run of the experiments.

RNN models

The script RNN_CV runs the RNN model on k-fold CV. It takes as arguments the following: - fold_dir: directory with the cross validation folds - outf: output file for the cross validation predictions - (opt) model_name: name for the output model - (opt) max_len: number of utterances to consider (between 500 and 1000, defaults to 500). - (opt) epochs: number of epochs to run the model (defaults to 30) - (opt) batch_size: batch size (defaults to 16) - (opt) FEAT: list of names of features (defaults to 'ngrams', 'w2v', 'afinn', 'vader', 'hatebase', 'empath_192', 'abusive', 'empath_2')