Problem
SemEval-2020 Task 12 was an AI competition held in 2020. The goal was to use AI to identify offensive language in social media.
![offensive language detection in social media](https://assets-global.website-files.com/655e4950d2a5437a1287a293/65f04b9ae1cc58d9367a4782_Group%201000004102.png)
The leaderboard of the competition is presented in the following table.
![](https://assets-global.website-files.com/655e4950d2a5437a1287a293/65f04bc75944fff283578ff6_Group%201000004225.png)
Approach
We used only the teacher-student paradigm to transfer knowledge from a large model (Teacher) to a small model (Student).
Result
iSemantics team surpassed 81 teams and scored 93.19% F1-score. We achieved this result using a model 3x smaller than the UHH-LT team, who ranked 1st.
A 3x smaller model means 3x faster decision time and 3x lower running cost when hosting on cloud services.
To learn more about automated data labeling techniques, read the full article here.