Example Comments Submission Form
Thanks for your interest in Perspective and the Conversation-AI research project (https://conversationai.github.io/)

This form allows you to contribute examples of comments to Google to be used for research and products to improve conversations online. This can include examples of comments that you want in your community, and examples of comments that may be hurting your community. If your examples includes more than 1 million rows, please contact: conversationai-questions@google.com, we'd love to discuss how to collaborate.


We can take the data in many formats (CSV, JSON, even a spreadsheet), so don’t worry about format too much, but the ideal data format (particularly for large datasets) is as a newline delimited JSON file (each line is a valid JSON entry info at ndjson.org).

For example, you can submit a file "comments.json" with the following fields:

FIELD DESCRIPTION
source---------------------------------------name of source/publication/section
comment_id------------------------------unique id for the comment
article_id-----------------------------------unique id for the article
comment_author_id------------------anonymized unique id for commenter
parent_id-----------------------------------null, or comment_id of parent in thread
comment_text---------------------------text of the comment
labels-----------------------------------------list of strings of labels associated with this comment;

Labels can include rejection reasons like "toxic", "off-topic"; positive labels, e.g. "editors-pick" etc; user-flags, e.g. "flagged-as-spam-by-user"; or anything else you think adds information that could help make a conversation better by knowing.

All fields except "comment_text" and "labels" are optional; we need those so that we can understand what the comments are, and what they are examples of.

As JSON this would look something like so:

{ “source”: “politics chat", “article_id”: "90163", “comment_author_id: "4acf39f1e2", parent_id: "47210", comment_text: "You are a stupid idiot", comment_id: "47212", labels: ["obscene"]}
{ “source”: “politics chat", “article_id”: "90163", “comment_author_id: "e9af5bb45", parent_id: "47212", comment_text: "You, are the real dummy here! fool!", comment_id: "47213", labels: ["personal_attack"]}

It's helpful to have context for the comments too, e.g. the article the comment is on, which you can also send us, for example in a file "articles.json" with fields something like this:

FIELD DESCRIPTION
source--------------------------name of source/publication/section
article_id-----------------------unique id for the article
article_title--------------------title of article
article_text--------------------text of the article
author---------------------------name of the article author (helps identify comments that attack the author)

Email address
Name
Your answer
Organization
Your answer
Description of the examples
Number of comments, description of fields, toxic vs non-toxic, language, etc.
Your answer
Share the data via Google Drive
Instructions: 1) Upload file to your Google Drive 2) Share the file with perspective-data@google.com 3) Copy link to the file and paste below.
Your answer
Sharing and Terms
Required
May we share these examples under a Creative Commons license as a public research resource?
Examples of existing public datasets can be found at conversationai.github.io
Submit
Never submit passwords through Google Forms.
This form was created inside of Google.com. Google - Privacy & Terms - About Google