ABOUT THE CHALLENGE:

This challenge is organized by the Data Science Club at Indiana University(DSC@IU).

REGISTRATION | TEAM CREATION & SUBMISSION:

In order to make submissions, please follow the below instructions

Register for the challenge, and create your team here. Only one person from a team must create a team and your team.
Once you create the team, you can invite the rest of your team members.
You will be able to see a submit button where you submit your prediction files and notebooks.

PRIZES AND GUIDELINES:

All teams need to upload there Jupyter Notebooks with the following details:
- Code with well commented/explained steps and approach
- Link to video presentation - end to end explanation of the project by one or more team members.
Teams with commendable performance will be selected for the evaluation.
Top 2 teams will be considered for potential RA opportunities in addition to winning upto 200 dollars prize money.

PROBLEM STATEMENT:

You are given:

A) A labeled verified dataset (“Gold Standard”) of about 4,000 tweets. These tweets have been labeled antisemitic/ not antisemitic by expert annotators who went over the same tweets and discussed any disagreements. Tweets were taken out of the dataset if no agreement was found. The tweets in the dataset are from January 2019 to April 2021. The were scraped with one of the four keywords “Jews,” “Israel,” “kikes,” or “ZionNazi*.”

B) 500 tweets that were labeled antisemitic/ not antisemitic by a group of students (“lay annoytators”). The tweets in this dataset are from January to April 2021 and were scraped with the keyword “Jews.” The same tweets were also classified by the expert annotators and are part of the Gold Standard.

TASKS:

Using data from Gold Standard (data from A)

Find a model that can predict antisemitic/ non-antisemitic tweets within the gold standard (take 80 or 90 percent of the data and run your test model on the remaining 20 or 10 percent). You might want to run your model separately for the tweets with generic keywords (“Jews” and “Israel”) and for the tweets with slur keywords (“kikes” and ZioNazi*”). Generate accuracy and efficiency results.
Use name recognition tools for information extraction to identify prominent categories such as geopolitical entities, person names, organisations and locations in the antisemitic and in the non-antisemitic tweets.
Examine the gold standard for linguistic radicalization patterns. At what periods do derogatory and pejorative terms and slurs appear?
Visualize differences between antisemitic and non-antisemitic tweets.

Using data from the Gold Standard and the lay annotations (data from A and B)

Use your model to predict the classification of tweets from dataset B and compare that in a meaningful way to the classification by the lay annotators (classification by students).
Predict the expert classification by the lay annotations. You might want to include labels other than yes/no antisemitic (the annotation also included other labels, such as sentiment, related to the Holocaust, Calling Out).

EVALUATION:

Submissions are evaluated using Accuracy Score.

How do we do it?

We give participants two sets of data:
- Training Dataset: It contains both independent and dependent variables. Anyone can download it, build a model and make a submission.
- Testing Dataset: It contains only independent variables (the dependent variables are hidden). You submit the prediction values of the dependent variables for this dataset and we compare it against the actual values.

Tweet Classification

Challenge Starts

Registration Ends

Challenge Ends