Some problems faced in Sentiment Analysis are:
- Sarcasm and Irony: When people use positive words to express negative sentiment the model can not predict it. This can be solved by using specific sarcasm detector algorithms.
- Types of Negation: When a sentence uses negation to reverse the polarity of the words or sentences the models find it hard to detect it if the word used to negate is not close to the subject. Example: "I do not like to study this subject," the effect of 'not' is only felt in the end of the sentence.
- Multipolarity: A single text can have many different sentiments (both positive and negative), so taking the mean of the text as a whole will not give the most accurate sentiment of the whole text.