Emotion_Analysis_Twitter

Code for:

Understanding the Role of Affect Dimensions in DetectingEmotions from Tweets: A Multi-task Approach (Submitted to SIGIR, Under Review)

Setting up:

pip install -r requirements.txt

Preparing dataset:

add the train, val and test csv files to data/ folder

How to run:

For scraping/hydrating:

python scrape.py -s True -q [queries] -l [limit on tweets]  
python scrape.py -H True -f [files containing tweets ids]

Note : The -H stands for hydration, and -s for scraping. Restrictions related to coordinates, time intervals, can be modified directly in the script.

For training :

python train.py --exp_name (value) --encoder (value) --data_dir (value) --save_dir (value) --lr (value) --batch_size (value) --save_policy (value) --activation (value) --optim (value) --wd (value) --epochs (value) --seed (value) --use_gpu(to use gpu) --use_empath(to use empath) --l2(to use l2 reg.) --use_scheduler(to use sched) --use_dropout(to use dropout)

For generating predictions :

python generate_predictions.py --gpu_id (gpu_id) --model_name (BERT/ROBERTA) --model_path (path to saved model) --output_path (path to save dir) --data (path to dir containing hydrated csv) --use_empath (y/n) --activation (tanh/bce)

For plotting graphs :

python plot.py -f (csv_or_tsv_file_with_predictions) -i(unique_id_field) -d(date_field) -e(field_with_emotion_predictions) -b(text_field) -l(boolean_flag_for_leap_year) -t(timestep_for_type_2) -c(chunk_size_for_type_1) -a(address_of_aspect_file)
Note-1: Plot types: (1)Fixed number of tweets, (2)Fixed time interval, (3)Aspect mentions (for fixed number of tweets out of the total)  
Note-2: By default the aspect term searching is case sensitive
Note-3: Aspects labels can be supplied in the code

New flags:

• --save_pickle ; to create dataset_{ENCODER}.pkl file
• --load_pickle PATH ; path to pickle file
• --job_type ; mtl, stl_emotion, stl_VAD
• --final_test ; boolean flag to combine train and val set to generate final test

Saving scheme:

• final_test=false, stores best validation epoch
• final_test=true, stores test result after training on best validation hp.json