The influence of sentiment polarization and exchange in online social networks has been growing and studied by many researchers and organizations worldwide. For example, the sentiments expressed in a text concerning a topic in the discussion tend to influence a community when a Twitter user retweets the original text, causing a chain of reactions within a network. This paper investigates sentiment polarization in Twitter, focusing on tweets with the hashtags #Coronavirus, #ClimateChange, #Immigrants, and #MeToo. Specifically, we collect the tweets mentioned above and classify them into five categories: hate speech, offensive, sexism, positive, and neutral. In this context, we address the problem as a multiclass classification problem by using the pre-trained language models ULMFiT and AWD-LSTM, which achieved a Fmicro of 0.85. Finally, we use the classified dataset to conduct a case study in which we capture the sentiment orientation during the network evolution.