NCA-GENM Question 120
Select 3You are tasked with designing a customer support chatbot for an e-commerce platform using a large language model (LLM). To improve the chatbot's ability to provide accurate and up-to-date responses about product availability, you decide to implement a retrieval-augmented generation (RAG) approach. Which of the following steps are essential to building this solution?
- A
Use a vector database to store and retrieve product information based on semantic similarity.
- B
Fine-tune the LLM with a dataset containing only historical customer queries and responses.
- C
Integrate a document retriever module to extract relevant product data during user interactions.
- D
Pre-train the LLM on the entire e-commerce platform’s product catalog.
- E
Combine retrieved product information with the LLM’s generative capabilities to compose responses.
Show answer and explanation
Correct answers: A, C, E
Explanation
The retrieval-augmented generation (RAG) approach enhances LLM capabilities by combining generative text generation with real-time retrieval of external information. In this scenario, using a vector database, a document retriever, and leveraging the LLM’s generative abilities are essential steps to ensure the chatbot provides accurate, up-to-date, and contextually relevant answers about product availability.
- A. Correct.
Correct: A vector database is commonly used in RAG systems to store and retrieve information efficiently based on semantic similarity, which is crucial for responding to customer inquiries about products.
- B. Incorrect.
Incorrect: Fine-tuning with historical queries alone may not address the need for real-time or up-to-date product data, which is a key requirement in this scenario.
- C. Correct.
Correct: A document retriever is necessary for fetching relevant product data during interactions, ensuring the chatbot has access to accurate and contextual information.
- D. Incorrect.
Incorrect: Pre-training the LLM on the entire product catalog is not practical or necessary for a RAG approach, as it focuses on retrieval rather than storing all data in the model itself.
- E. Correct.
Correct: Combining retrieved information with the LLM’s generative capabilities is a core principle of RAG, enabling the chatbot to generate coherent and informative responses.