Dataset Card for [Dataset Name]

Dataset Summary

Comments from Replit's Community, sourced via moderator GraphQL queries and personally labeled :). For use in Replit + Weights and Biases Hackathon.

Supported Tasks and Leaderboards

Text Classification

Languages

English

Dataset Structure

Data Instances

{"label":3,"text":"@KENDALPETERSON\nShut up you dont have a permit to brag."}

Labels

0: General
1: Spam
2: NSFW
3: Harassment

Data Fields

Label, Text

Data Splits

Train, Validation, Test

Dataset Creation

Curation Rationale

Fine-tuning data for the Replit + Weights and Biases Hackathon.

Source Data

Initial Data Collection and Normalization

This data was collected via Replit's GraphQL API using a query only available to site moderators and admins, allowing for querying comments by pattern. By not setting a pattern one could get up to 50 comments within a given time span. Data was sourced from February 1st, 2023 to February 10th, 2023.

Who are the source language producers?

Replit Community users

Annotations

Annotation process

I tagged the data using a small website that I made to assign labels to individual comments.

Who are the annotators?

Me (@rayhanadev)

Personal and Sensitive Information

This dataset is purely unfiltered and may contain personal and sensitive information.

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

The data is categorized according to three categories of content that are not allowed on Replit (and would fall into a fourth "general" category if it is acceptable content).

Other Known Limitations

The dataset is small (n<1k) but I am working on increasing the amount of available data. There is also significantly more unacceptable content labelled than there is acceptable ("general") content.

Additional Information

Dataset Curators