task list - Githubissues

kbkreddy / Hate-Speech-and-Offensive-Language-Identification

Project on the HASOC-Offensive Language Identification FIRE 2020 tasks, which includes a message-level classification task that classifying a twitter comment into the offensive (OFF) or Not-offensive (NOT) language. To achieve this goal, in this project, we propose an ensemble model which makes full use of the information of rich sequential patterns. More precisely, the proposed model contains combined model of BERT and SVM. Achieved the F1 values of 0.51 and ranked 5th.

1 stars 0 forks source link

task list #1

Open PJramya opened 3 years ago

PJramya commented 3 years ago

** Try out the below models and maintain separate clean notebooks for them

[x] Logistic Regression
[x] SVM
[x] LSTM
[x] using BERT

Languages to try:

[x] Hindi
[x] German

Analysis:

[x] Data cleaning
[ ] using Rules
[ ] score analysis

PJramya commented 3 years ago

[ ] trying SMOTE

PJramya commented 3 years ago

Latest scores page (macro avg):

English TASK1:

SVM: 0 0.80 0.91 0.85 391 1 0.90 0.78 0.84 423

LSTM: HOF 0.90 0.75 0.82 423 NOT 0.77 0.91 0.83 391

Hindi TASK1:

SVM: 0 0.76 0.86 0.81 466 1 0.52 0.37 0.43 197

LSTM: HOF 0.55 0.30 0.39 197 NOT 0.75 0.90 0.82 466