dataset_info:
features:
- name: text
dtype: string
- name: label
dtype: int64
- name: predicted_sentiment_facebook/bart-large-mnli
dtype: string
- name: predicted_sentiment_distilbert-base-uncased
dtype: string
- name: predicted_sentiment_roberta-base
dtype: string
splits:
- name: train
num_bytes: 1361555
num_examples: 1000
download_size: 862047
dataset_size: 1361555
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: odbl
task_categories:
- text-classification
language:
- en
size_categories:
- n<1K
Dataset Description
In this Task , we conducted one-shot sentiment analysis on a subset of the IMDb movie reviews dataset using multiple language models. The goal was to predict the sentiment (positive or negative) of movie reviews without fine-tuning the models on the specific task. We utilized three different pre-trained language models for zero-shot classification: BART-large, DistilBERT-base, and RoBERTa-base. For each model, we generated predicted sentiment labels for a subset of 100 movie reviews from the IMDb dataset. The reviews were randomly sampled, ensuring a diverse representation of sentiments. After processing the reviews through each model, we saved the predicted sentiment labels alongside the original reviews in a CSV file named "imdb_reviews_with_labels.csv". This file contains the reviews and the predicted sentiment labels for each model. Additionally, we uploaded both the dataset and the CSV file to the Hugging Face Hub for easy access and sharing. The dataset can be found at the following link after uploading:
https://huggingface.co/datasets/Mouwiya/imdb_reviews_with_labels
This task demonstrates the effectiveness of zero-shot classification using pre-trained language models for sentiment analysis tasks and provides a valuable resource for further analysis and experimentation.
- Curated by: [Mouwiya S. A. AlQaisieh]