I am experimenting with natural language processing for classification tasks and I need to train a model on a dateset that includes a mix of technical, gaming, and community terms such as roblaxmod alongside other domain text. I am finding that the model struggles with rare tokens and out-of-vocabulary terms which impacts accuracy. What preprocessing techniques or tokenization strategies have you found effective when dealing with mixed vocabulary datasets to improve model performance and maintain meaningful representations for uncommon terms Thanks for your insights.