Aggressive Language Identification Using Word Embeddings and Sentiment Features

This notebook presents the code of my participation in the First Shared Task on Aggression Identification. More details about the approach can be found in:

Constantin Orǎsan (2018) Aggressive Language Identification Using Word Embeddings and Sentiment Features. In Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018), p. 113 - 119, Santa Fe, USA, August, 25, http://aclweb.org/anthology/W18-4414

Requirements

In order to run this notebook you will need the following software:

• python3.6
• NLTK
• numpy and sklearn

Probably the easiest way to get all these is to install Anaconda.

You will also need to obtain the training, development and testing data from the workshop organisers and store the CSV files in a folder called english (or set the correct path in the path_to_data).

Download the GloVe vector from https://nlp.stanford.edu/projects/glove/. The current code assumes you use the 300d vectors. Uncompress the file and update the path_to_glove variable to point to it.

Uncompress the senti_feats.7z. This file contains the scores assigned by the SentiStrength. If you want to process different texts you will need to obtain SentiStrength from its author and preprocess the data using the procedure described in the notebook.