Sentiment Analysis Twitter Dataset

I want to do sentiment analysis using gpt-3.5-turbo by giving a prompt or message role: system, content: contains the provisions of each sentiment (positive, negative, neutral). role: user, content: tweet (this variable is obtained from reading the twitter dataset stored in csv and excel files). The expected result is only 1 token, namely positive then the result is 1, negative then the result is -1, and neutral then the result is 0.
I really hope for your help, thank you.