Aspect-Based Sentiment Analysis

This repository contains the code and data for the "Czech Dataset for Complex Aspect-Based Sentiment Analysis Tasks" accepted to LREC-COLING 2024 conference.

Requirements

Python 3.10 and higher is required. The code was tested on Python 3.11.4. The required Python packages are listed in the file requirements.txt. They can be installed using the command pip install -r requirements.txt.

Structure

The repository contains the following files and folders:

• main.py – Main file of the application.
• requirements.txt – List of required Python packages.
• README.md – This file.
• data – Folder containing the data. See the section Data for more details.
• src – Folder containing the source code.
  • data_utils – Folder containing the code for data loading and preprocessing.
  • evaluation – Folder containing the code for evaluation.
  • models – Folder containing the code for models.
  • utils – Folder containing the code for other utilities.

Data

There should be a folder data in the root of the repository. The folder should contain the following subfolders:

• cs_rest_o – Folder containing the data for CsRest-O dataset.
• cs_rest_n – Folder containing the data for CsRest-N dataset.
• cs_rest_m – Folder containing the data for CsRest-M dataset.

Each of the language folders should contain the following files:

• train.xlm – File with training data.
• dev.xlm – File with validation data.
• test.xlm – File with test data.

Usage

To run the code, use the command python main.py. The code can be configured using the following command-line arguments:

• --model – Name or path to pre-trained model. The default value is t5-base.
• --batch_size – Batch size. The default value is 64.
• --max_seq_length – Maximum sequence length. The default value is 256.
• --max_seq_length_label – Maximum sequence length for labels. The default value is 256.
• --lr – Learning rate. The default value is 1e-4.
• --epochs – Number of training epochs. The default value is 10.
• --dataset_path – Dataset path. The default value is cs_rest_m. Options: cs_rest_o, cs_rest_n, cs_rest_m.
• --optimizer – Optimizer. The default value is AdamW. Options: adafactor, AdamW. adafactor can be applied only for sequence-to-sequence models (e.g. T5), otherwise AdamW is always used.
• --mode – Mode of the training. The default value is dev. Options:
  • dev – Splits the training data into training and validation sets. The validation set is used for selecting the best model, which is then evaluated on the test set of the target language. In the case of different source and target languages, the validation set is taken from the training data of the target language and the whole training data is used for training.
  • test – Uses whole training dataset from the source language for training and test dataset from the target language for evaluation.
• --checkpoint_monitor – Metric based on which the best model will be stored according to the performance on validation data in dev mode. The default value is val_loss. Options: val_loss, ate_f1, acd_f1, acte_f1, tasd_f1, pd_f1, e2e_f1.
• --accumulate_grad_batches – Accumulates gradient batches. The default value is 1. It is used when there is insufficient memory for training for the required effective batch size.
• --beam_size – Beam size for beam search decoding. The default value is 1.
• --task – ABSA task. Options: ate, e2e, apd, acd.

Restrictions and details

• --task cannot be used for sequence-to-sequence models (e.g. T5).
• When selecting a model for sentiment polarity classification (e.g. BERT), --task has to be used.
• The user is responsible for selecting the correct --checkpoint_monitor (e.g. acd_f1 is only measured with acd task).
• The -model argument containing the substring t5 or bart indicates the use of a sequence-to-sequence model.
• The program automatically detects whether it is possible to use GPU for training.
• The program tries to use CsvLogger.

Examples

• python main.py --model google/mt5-base --batch_size 32 --max_seq_length 256 --max_seq_length_label 256 --lr 3e-4 --epochs 35 --dataset_path cs_rest_o --optimizer adafactor --mode dev --checkpoint_monitor tasd_f1 --accumulate_grad_batches 2 --beam_size 1
• python main.py --model xlm-roberta-base --batch_size 64 --max_seq_length 256 --lr 1e-5 --epochs 50 --dataset_path cs_rest_o --optimizer AdamW --mode dev --checkpoint_monitor pd_f1 --accumulate_grad_batches 1 --task apd
• python main.py --model xlm-roberta-base --batch_size 64 --max_seq_length 256 --max_seq_length_label 256 --lr 1e-5 --epochs 50 --dataset_path cs_rest_m --optimizer AdamW --mode test --task ate
• python main.py --model xlm-roberta-base --batch_size 64 --max_seq_length 256 --max_seq_length_label 256 --lr 1e-5 --epochs 10 --dataset_path cs_rest_n --checkpoint_monitor acd_f1 --optimizer AdamW --mode dev --task acd