iberbench/tass2020_sentiment_analysis

Dataset Card for TASS 2020 - Sentiment Analysis Task

Dataset Details

Dataset Description

The TASS 2020 dataset for Sentiment Analysis consists of Spanish-language social media posts annotated for sentiment polarity. The dataset includes examples labeled as positive, neutral, or negative, and incorporates regional variations of Spanish from different Spanish-speaking countries.

• Curated by: TASS 2020 Organizers and SEPLN
• Funded by: [Information Needed]
• Shared by: [Information Needed]
• Language(s) (NLP): Spanish (BCP-47: es)
• License: [Information Needed]

Dataset Sources

• Repository: TASS 2020 Repository
• Paper: TASS 2020 Overview
• Demo: [Information Needed]

Uses

Direct Use

This dataset is intended for training and evaluating models for sentiment polarity classification, where models classify text into positive, neutral, or negative sentiment categories.

Out-of-Scope Use

This dataset is not suitable for tasks requiring detailed context or analysis beyond sentiment polarity. Misuse includes drawing conclusions about individuals or demographics without considering biases in the dataset.

Dataset Structure

The dataset is divided into train and test splits. Each instance includes the following fields:

• text: The original Spanish-language social media post.
• label: The sentiment label (positive, neutral, or negative).
• language_variation: The regional variation of Spanish (e.g., Spain, Latin America).
• language: The language code (es for Spanish).

Split	Number of Instances	Size (Bytes)
Train	10,000	1,052,432
Test	5,000	645,219

Dataset Creation

Curation Rationale

The dataset was created to provide a benchmark for sentiment analysis in Spanish, reflecting real-world challenges with informal language on social media platforms.

Source Data

Data Collection and Processing

The data was collected from public social media platforms and filtered to remove personal information. Annotation guidelines were provided to ensure consistency, and data normalization addressed spelling errors and informal expressions.

Who are the source data producers?

The data originates from Spanish-speaking users on social media platforms. No demographic information is available.

Annotations

Annotation process

Annotations were performed by Spanish-speaking NLP experts and linguists who labeled each post with its sentiment polarity.

Who are the annotators?

Annotations were created by a team of trained annotators with expertise in sentiment analysis.

Personal and Sensitive Information

The dataset has been anonymized, with no personal identifiers retained.

Bias, Risks, and Limitations

The dataset reflects biases inherent in social media content, such as skewed sentiment distributions or underrepresentation of certain demographic groups. Informal language and regional variations may affect generalizability.

Recommendations

Users should consider pre-processing steps and potential biases when using this dataset for downstream tasks. Models trained on this dataset may not generalize to formal or non-social media text.

Citation

BibTeX:

@inproceedings{tass2020_sentiment_analysis,
    title = "{TASS} 2020: Sentiment Analysis in Spanish Social Media Text",
    author = "Villena-Román, Julio  and Correa, Elizabeth",
    booktitle = "Proceedings of the TASS 2020 Workshop",
    year = "2020",
    address = "Alicante, Spain",
    publisher = "SEPLN",
    url = "http://tass.sepln.org/2020/"
}