A brief introduction to the task and the dataset used

• I used this dataset from kaggle
• It has 4 categories World, Sports, Business ,Science/Technology

The preprocessing steps taken

• Used DistilBertTokenizer to tokenize the sentences

The architecture of the model used, and how it was fine-tuned

• I used distilbert-based-uncased model form huggingface and added two linear layers and one dropout layer for classification task

The evaluation metrics and the results obtained

• I used cross entrophy loss to calculate the loss
• After training 1 epoch with whole dataset and batch size of 16, the accuracy of model was 83.88% which is preety good for four classes and if i had trained for 5 epochs it could have reached 90% easily.

A discussion of the performance of the model and possible ways to improve it.

• I didnt had enough time to train the model so i was only able to get 83.88% accuracy, it could be improved by training for atleast 5 hours.