Generative AI

Prompt Engineering for NLP tasks
Sharath Gilla
June 12, 2023

In the world of artificial intelligence, the advent of language models like ChatGPT has revolutionized the way we interact with technology. These models are designed to understand and generate human-like text, opening endless possibilities for applications such as chatbots, virtual assistants, and content generation. One key technique that has emerged as a powerful tool in enhancing the capabilities of language models is prompt engineering. In this blog post, we will explore the concept of prompt engineering and its significant role in maximizing the potential of ChatGPT.

What does the term "prompt" mean?

A prompt is a piece of text provided as input to the model, which helps to generate responses. Prompt engineering involves carefully crafting prompts which is the key to achieve desired outcomes, influence the model's behaviour, or encourage specific types of responses.

By utilizing this technique effectively, developers can shape the interactions with ChatGPT, making it more reliable, coherent, and aligned with user intentions.

By fine-tuning ChatGPT on sentiment analysis tasks, it can be used as a classifier, directly predicting a given text’s sentiment.

Leveraging the power of prompt engineering:

  • Improved Control: By providing explicit instructions or constraints in the prompts, it is possible to steer the model towards generating desired content, ensuring more accurate and relevant outputs.
  • Bias Mitigation: Language models can inadvertently generate biased or sensitive content based on the data they are trained on. Prompt engineering can be employed to reduce such biases by incorporating guidelines or instructions that promote fairness, inclusivity, and neutrality in the generated responses.
  • Content Style and Tone: With prompt engineering, it is possible to fine-tune ChatGPT's output to match specific writing styles or tones. Whether it's a professional tone for business communication or a casual style for friendly interactions, prompts can guide the model to generate text that aligns with the desired tone.
  • Error Correction and Clarification: ChatGPT may sometimes produce inaccurate or ambiguous responses. Prompt engineering allows developers to include clarifying questions, disambiguating statements, or reference materials to prompt the model to provide more accurate and helpful answers.
  • Domain Adaptation: By crafting prompts that include domain-specific information, developers can enhance ChatGPT's understanding of specialized topics. This enables the model to provide more informed and accurate responses in various fields like medicine, law, or technology.  

Sentiment Analysis using Prompts:  

In the context of language models like ChatGPT, sentiment analysis is a process of evaluating the sentiment or emotional tone of a given sentence, text, or document. It involves examining the words, phrases, and overall linguistic cues present in the text to classify it into categories such as positive, negative, or neutral.  

To perform sentiment analysis, the language model analyses various linguistic features, including the presence of positive or negative words, intensity of emotion, grammatical structure, word order, context, and overall sentiment indicators. These features help the model understand the sentiment conveyed in the text.  

The classification into positive, negative, or neutral categories allows the language model to provide a quantitative representation of the sentiment expressed in the text. This information can be useful in many applications, such as social media monitoring, customer feedback analysis, market research, and brand reputation management.  

It's important to note that sentiment analysis is not always a straightforward task, as the interpretation of sentiment can vary based on the context and subjective understanding. Language models like ChatGPT aim to capture the sentiment accurately but may still encounter challenges in handling sarcasm, irony, or complex emotional expressions.

OpenAI models are trained on vast amounts of data from various sources, which may include biased or problematic content. Users have limited control over the specific data used for training, which can lead to potential biases in sentiment analysis and non-reliable answers.

Below is the example where ChatGPT struggles to give the sentiment when we use different prompts.  

The modification of the prompt used to ascertain the sentiment in the sentence resulted in a decrease in ChatGPT's confidence level regarding the provided statement.

The prompt-[text] analyse the overall sentiment in this review. treats the text differently and provides different prediction compared to the prompt-classify the sentiment expressed in the following: [text].

Here is another example for the similar one.

In the below image it cannot accurately classify the text into sarcasm, instead gives Neutral answer.

But, in few cases it identifies the sarcasm as well as frustration.

From above observations these are the few cons of ChatGPT in this use case.

  • Contextual understanding: The performance of sentiment analysis heavily relies on the context provided in the conversation. If the context is ambiguous or incomplete, it may lead to inaccurate results.
  • Lack of domain: Specific knowledge: Sentiment analysis models, like ChatGPT, have a general understanding of language but may lack specialized knowledge in specific domains. This can affect the accuracy of sentiment analysis in industry-specific or technical conversations that require domain-specific terminology and nuances.
  • Inability to detect sarcasm or irony: Sentiment analysis models, including ChatGPT, may struggle to accurately detect sarcasm, irony, or subtle nuances in sentiment. This can result in misinterpretation and misclassification of sentiment when the text contains such linguistic elements.


Prompt engineering is a powerful technique that allows developers to leverage the capabilities of ChatGPT to its fullest potential. By carefully crafting prompts, we can enhance control, mitigate biases, shape content style, and improve the overall accuracy of the model's responses. With continued research and exploration in this field, prompt engineering holds the key to unlocking more refined and reliable interactions with language models, paving the way for more effective and engaging AI-powered applications.

We hope you found our blog post informative. If you have any project inquiries or would like to discuss your data and analytics needs, please don't hesitate to contact us at We're here to help! Thank you for reading.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.