Google

An application of transforming NLP problems into text generation involves recasting text classification. Instead of training a model to output a specific class label, a large language model can be prompted to generate text that explicitly states the classification, thereby converting the classification problem into a generation task.

Example of Reframing Text Classification as Text Generation

A simple prompting method for classification involves concatenating the input text with a cue phrase to form a complete prompt. For example, the input 'I love the food here. It’s amazing!' can be combined with 'I’m' to create the prompt: 'I love the food here. It’s amazing! I’m'. The model's generated completion is then used to determine the classification.

Example of a Prompt for Classification via Completion

Based on the provided scenario, formulate a complete instruction (a prompt) that would guide a large language model to classify the user review by generating a sentence or a descriptive phrase, rather than just a single-word label. 

**User Review:** 'The movie was great, but it kept buffering every five minutes on my device.'

Reframing Review Classification

A developer needs to classify short user feedback messages into one of three categories: 'Bug Report', 'Feature Request', or 'General Inquiry'. Following the principle of reframing classification as a generation task, write a complete and effective prompt you would give to a large language model to classify the following user message: 'The login button isn't working on the mobile app.'

Prompt Design for Generative Classification

A developer is building a system to categorize user reviews as either 'Positive' or 'Negative'. A traditional approach would involve a model that outputs a single, predefined label (e.g., the word 'Positive'). How does reframing this task as a text generation problem for a large language model fundamentally change the model's expected output?

Learn Before

Related