C-L-V/mini_corpus_sentiment_en_translated

Manual Mini Corpus for Sentiment Analysis

Dataset Description

This dataset consists of a small corpus of 50 texts manually collected from the social media platform X (formerly Twitter). Each text has been manually annotated for a sentiment analysis task using three categories: positive, negative, and neutral. The distribution across the three categories is as follows: 19 positive, 19 negative, and 12 neutral instances.

The original texts were collected in Spanish and subsequently translated into English using :contentReference[oaicite:0]{index=0}. The translations were manually reviewed to ensure that they preserve, as faithfully as possible, the meaning and sentiment of the original texts.

All posts were collected in mid-April 2026.

---

Dataset Structure

The dataset is provided in CSV format with the following columns:

• id: numerical identifier of the text
• text: content of the text
• label: manually assigned sentiment label

---

Labels

The sentiment labels used in this corpus are defined as follows:

• positive: the text expresses a favorable evaluation
• negative: the text expresses an unfavorable evaluation
• neutral: the text conveys an ambiguous or mixed emotional load, or is primarily informational

---

Dataset Creation

Curation Rationale

This dataset was created for academic purposes to support experimentation in sentiment analysis, particularly in low-resource and small-scale evaluation settings.

Source Data

The texts were manually collected from publicly available posts on X (Twitter). No private data has been used.

Annotations

All texts were manually annotated by the dataset creator. The annotation process follows a simple three-class sentiment scheme (positive, negative, neutral).

---

Intended Use

This dataset is intended for:

• Evaluating sentiment analysis models
• Comparing different approaches (e.g., machine learning, deep learning, transfer learning, or LLM-based methods)
• Serving as a small benchmark or illustrative dataset in academic contexts

---

Limitations

• The dataset is very small (50 examples) and therefore not representative of general language use
• Manual annotation introduces a degree of subjectivity
• The texts were automatically translated and then manually reviewed, but some nuances from the original Spanish may not be fully preserved

---

Bias and Risks

Due to its limited size and manual curation, the dataset may reflect unintended biases from both the selection of texts and the annotation process.

---