Using Prompts to Implement Intent Alignment in an Intelligent Customer Service System

In an intelligent customer service system, the LLM pairs a user query with an FAQ title that shares the same meaning and returns the title. Using this title, the system locates a question-answer (QA) pair and uses it to respond to the user query.

Step 1: The task description is unclear, and the model's answer is incorrect.

Prompt:In an intelligent Q&A scenario, a customer asks "How do I deregister my credit card?" Which of the following FAQ titles should be returned? Options: How do I bind a bank card? How do I delete a bank card?Model answer:How do I deregister a bank card?

Step 2: Provide the background information, detailed task description, and rules. The model's answer is correct.

Prompt:Task requirements:You must have a solid understanding of financial knowledge as well as the true intent behind both the customer's question and the available standard question options. Based on the provided question and standard options, select the standard question that best matches the customer's question. You can only select from the standard question options. Do not add, remove, or modify the options. Return only the matching result. No other information needs to be returned.Task start:Customer's question: How do I cancel my credit card?Standard question options: How do I bind a bank card? How do I remove a bank card?Model answer:How do I remove a bank card?

Step 3: If the customer question does not match the standard question, the model will fabricate an option.

Task requirements:You must have a solid understanding of financial knowledge as well as the true intent behind both the customer's question and the available standard question options. Based on the provided question and standard options, select the standard question that best matches the customer's question. You can only select from the standard question options. Do not add, remove, or modify the options. Return only the matching result. No other information needs to be returned.Task start:Customer's question: How do I change my credit card?Standard question options: How do I bind a bank card? How do I remove a bank card?Model answer:In this example, the standard question that best matches the customer's question is "How do I change my credit card?"

Step 4: List the specific requirements and the corresponding strategy for handling mismatched situations. The model's answer is correct.

Task requirements:You must have a solid understanding of financial knowledge as well as the true intent behind both the customer's question and the available standard question options. Based on the provided question and standard options, select the standard question that best matches the customer's question. Requirements:1. Return only the standard question option. If the customer's question does not match any of the given standard question options, return "N/A".2. Only select from the standard question options. Do not add, remove, or modify the options. Return only the matching result.Task start:Customer's question: How do I change my credit card?Standard question options: How do I bind a bank card? How do I remove a bank card?Model answer:N/A