Modeling Affect Intensity

Modeling affect intensity expressed in tweets. Semeval task 2018. Authors - Aarsh Patel & Lynn Samson

Project Organization

├── LICENSE
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── dev            <- Dev data.
│   ├── train          <- Train data
│   ├── lexicons       <- Affect/Sentiment Lexicons
│   └── word2vec       <- Word2Vec/Glove embeddings.
│
├── models             <- Implementation of various models
│
├── notebooks          <- Jupyter notebooks
│
├── references         <- Relevant literatue for the project
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── utils.             <- Folder contain utility scripts (generating features, preprocessing tweets...)
|
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
├── evaluate.py        <- Evaluation script

Training and evaluating models

python evaluate.py --model model_name --features feature_type --metrics metric1, metric2 --optimize True/False

Baseline Model

We used lexicon features in addition to glove embeddings as features to a SVR model as our baseline model. In order to run the baseline model, run the following command:

python evaluate.py --model baseline

Results:

Emotion	CV Pearson Correlation Score
Anger	0.66019879032626883
Fear	0.67025025487576528
Joy	0.68789131129906456
Sadness	0.66106523454135879
AVG	0.66985139776061442

Best Performing Model

A simple feedforward neural network that uses word2vec embeddings and lexicon features as input into the model.

Results:

Emotion	CV Pearson Correlation Score
Anger	0.66019879032626883
Fear	0.67025025487576528
Joy	0.68789131129906456
Sadness	0.66106523454135879
AVG	0.66985139776061442

Experiments:

TODO