The GoFundMe dataset is part of the training data for the CommonFrame project. It is a collection of the individual fund raising campaigns available through the GoFundMe website. The dataset was designed to inform machine learning technologies about effective communications and campaign strategies to maximize stakeholder objectives.

Below is a list of files reflecting the entire process of mining the intial data, preprocessing raw text files, and transforming the resulting dataframe into usable information.

• combine_df.py : Combine csv files
• data_cleaning.ipynb : Initial data processing
• gofund_links_scraper.py : Script for links
• gofund_stats.ipynb : Data analysis with Pandas
• is_unique.py : Unique campaign urls
• profile.ipynb : Profile writeup
• scrape_by_campaign.py : Data miner
• sent_tokens.ipynb : Test tokenization
• txt_to_csv.py : Write out txt to csv

The GoFundMe dataset was collected using Python web technologies and is stored in csv format. Each campaign category from the GoFundMe website was loaded until all active campaigns were cached. The resulting urls were then stored in a separate file, where each url representing individual campaigns was loaded onto a custom built webscraper that targeted elements relevant to the project. The resulting html object is then analyzed to return only the relevant information for the project. The resulting csv file is then stored to disk and imported as a dataframe where further data processing is conducted. The final dataset is again stored in csv format and passed on to the CommonFrame database for further annotation by researchers. This forms the initial training data for machine learning models.

A total of 7077 campaigns was gathered under 18 different campaign categories. The data includes the individual titles and text description of the campaigns. Descriptive elements for individual campaigns were also included during the scrape which can be used for exploratory data analysis (i.e. EDA / descriptive statistics).

The dollar amount for each individual donation is not itemized – just the total sum of donations per campaign.

Please note that no attempt was made to verify the legitimacy of campaigns that were included in the dataset.

Deactivated campaigns were dropped from the collection.

The column story_tokens is a tokenized copy of the header: c_story (campaign descriptions). The Punkt sentence tokenizer from the Python Natural Language Toolkit was used to subset the campaign descriptions.

Below is a quick summary of the different campaign categories within the dataset. The first table shows the total number of campaigns per category. The second table compares the median number of donors and amount raised per category.