At the end of the day, classification accuracy depends directly on your feature engineering. Specifically looking at text classification problem, you can try using better word embedding for your text like Word2Vec, Glove, and Elmo. This article can help you with embeddings.
You can also look at pre-processing the text data correctly, for example, does the data has stop words or common words in all documents that do not add much info? you can remove these words and see how it impacts your classifier. Article for preprocessing.
And finally, the performance also depends upon the kind of ML model that you are using. Text classification approaches(ML) can help you here.