EconComments-RuBERT-sentiment

This model is a fine-tuned version of ai-forever/ruBert-base on a dataset of 7,000 Russian-language comments on economic topics from the VKontakte social network

Task

For comments on economic topics from the social network, the task of multiclass sentiment classification into 3 classes (negative, neutral, positive) was solved

• Quadratic Weighted Kappa: 0.532

Training hyperparameters

• learning_rate: 1e-05
• weight_decay: 0.01
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• num_epochs: 2