---
task_categories:
- text-classification
- feature-extraction
language:
- en
---
# Twitter Sentiment Meta-Analysis Dataset

## Dataset Description

This dataset contains sentiment analysis results for English tweets collected between September 2009 and January 2010. The tweets were processed and analyzed using 10 different sentiment classifiers, with the final sentiment score derived from principal component analysis (PCA).

## Source Data

- **Original Data**: Cheng-Caverlee-Lee Twitter Scrape (Sept 2009 - Jan 2010)
- **Number of Tweets**: 138 690
- **Language**: English only (filtered using langdetect)

## Data Processing

1. Text normalization:
   - Normalized bullet points, quotation marks, Unicode characters, and whitespace
   - Replaced emails, hashtags, phone numbers, URLs, and user handles with placeholder tokens

2. Language filtering:
   - Retained only English tweets using langdetect

3. Sentiment analysis:
   - Applied 10 different sentiment classifiers (see Models section)
   - Sentiment scores represent the difference between positive and negative logits, except for one model using a 5-star rating system

4. Final sentiment score:
   - Performed PCA on the 10 classifier scores (column scaling only, no centering)
   - Transformed PC1 to z-scores to obtain the final sentiment score

5. Training subset:
   - The sentiment scores were split into ten groups and 10000 from each group were used to make the training subset (72 415 rows)

## Models

The following sentiment analysis models were used:

1. [cardiffnlp/twitter-roberta-base-sentiment-latest](https://huggingface.co/cardiffnlp/twitter-roberta-base-sentiment-latest)
2. [lxyuan/distilbert-base-multilingual-cased-sentiments-student](https://huggingface.co/lxyuan/distilbert-base-multilingual-cased-sentiments-student)
3. [ahmedrachid/FinancialBERT-Sentiment-Analysis](https://huggingface.co/ahmedrachid/FinancialBERT-Sentiment-Analysis)
4. [finiteautomata/bertweet-base-sentiment-analysis](https://huggingface.co/finiteautomata/bertweet-base-sentiment-analysis)
5. [nlptown/bert-base-multilingual-uncased-sentiment](https://huggingface.co/nlptown/bert-base-multilingual-uncased-sentiment)
6. [siebert/sentiment-roberta-large-english](https://huggingface.co/siebert/sentiment-roberta-large-english)
7. [soleimanian/financial-roberta-large-sentiment](https://huggingface.co/soleimanian/financial-roberta-large-sentiment)
8. [FinanceInc/auditor_sentiment_finetuned](https://huggingface.co/FinanceInc/auditor_sentiment_finetuned)
9. [Kaludi/Reviews-Sentiment-Analysis](https://huggingface.co/Kaludi/Reviews-Sentiment-Analysis)
10. [StephanAkkerman/FinTwitBERT-sentiment](https://huggingface.co/StephanAkkerman/FinTwitBERT-sentiment)

## Features

- `id`: A unique ID for each tweet (note: not the same IDs as the original dataset)
- `text`: Original tweet text after processing
- `*-sentiment`: Individual sentiment scores from 10 classifiers
- `sentiment`: Final sentiment score (PCA-derived)

## Usage

This dataset is suitable for:
- Comparative analysis of sentiment classifiers
- Meta-analysis of Twitter sentiment
- Temporal analysis of public sentiment (Sept 2009 - Jan 2010)

## Limitations

- Limited to English tweets only
- Temporal scope limited to Sept 2009 - Jan 2010
- User names, hash tags, URLs, conversation IDs were removed in order to focus on sentiment
- Sentiment scores are model-dependent and may not reflect ground truth
- Lack of context in conversation may impact sentiment scores

## Citation

The original tweets are from:

Z. Cheng, J. Caverlee, and K. Lee. [You Are Where You Tweet: A Content-Based Approach to Geo-locating Twitter Users.](https://archive.org/details/twitter_cikm_2010) In Proceeding of the 19th ACM Conference on Information and Knowledge Management (CIKM), Toronto, Oct 2010.

All other analyses performed are my own.