fintwit-images / README.md
StephanAkkerman's picture
new data upload
945c7df verified
metadata
language:
  - en
license: mit
task_categories:
  - image-classification
  - image-feature-extraction
pretty_name: FinTwit Images
dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype:
        class_label:
          names:
            '0': charts
            '1': non-charts
    - name: id
      dtype: string
  splits:
    - name: train
      num_bytes: 320924227.116
      num_examples: 4177
  download_size: 444164916
  dataset_size: 320924227.116
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
tags:
  - fintwit
  - twitter
  - charts
  - financial
  - financial charts
  - finance
  - stocks
  - crypto
  - image

FinTwit Images

This dataset is a collection of a sample of images from tweets that I scraped using my Discord bot that keeps track of financial influencers on Twitter. The data consists of images that were part of tweets that did not mention a ticker. This dataset can be used for a wide variety of tasks, such as image classification or feature extraction.

FinTwit Charts Collection

This dataset is part of a larger collection of datasets, scraped from Twitter and labeled by a human (me). Below is the list of related datasets.

  • Crypto Charts: Images of financial charts of cryptocurrencies
  • Stock Charts: Images of financial charts of stocks
  • FinTwit Images: Images that had no clear description, this contains a lot of non-chart images

Dataset Structure

Each images in the dataset is structured as follows:

  • Image: The image of the tweet, this can be of varying dimensions.
  • Label: A numerical label indicating the category of the image, with '1' for charts, and '0' for non-charts.

Dataset Size

The dataset comprises 4,579 images in total, categorized into:

  • 1,083 chart images
  • 3,496 non-chart images

Usage

I used this dataset for training my chart-recognizer model for classifying if an image is a chart or not.

Acknowledgments

We extend our heartfelt gratitude to all the authors of the original tweets.

License

This dataset is made available under the MIT license, adhering to the licensing terms of the original datasets.