Databricks Generative AI Engineer Associate Question 53
Select 3You are developing a Retrieval-Augmented Generation (RAG) application using Databricks. The source documents contain a mix of product descriptions, customer reviews, and unrelated advertisements. The RAG application is generating irrelevant responses due to noisy content. What steps can you take to improve the relevance of the responses?
- A
Preprocess the documents to remove unrelated advertisements before indexing them.
- B
Use a text summarization model to condense the source documents before retrieval.
- C
Filter the source documents based on topic modeling to ensure only relevant content is used.
- D
Increase the temperature parameter in the language model's generation step to allow greater creativity.
- E
Leverage embeddings to rank and filter out content that is not semantically similar to the query.
Show answer and explanation
Correct answers: A, C, E
Explanation
In a RAG application, irrelevant or noisy content in source documents can degrade response quality. To improve relevance, it is crucial to preprocess the documents by removing extraneous content, apply filtering techniques like topic modeling, and leverage embeddings for semantic similarity-based ranking. These steps ensure that the RAG system retrieves high-quality, relevant information for generating accurate and meaningful responses.
- A. Correct.
Removing unrelated advertisements ensures that non-relevant content does not interfere with the retrieval and response generation process.
- B. Incorrect.
While summarizing documents can reduce their size, it does not specifically address the issue of removing extraneous, irrelevant content.
- C. Correct.
Topic modeling can help identify and filter out content that is unrelated to the desired domain or query, improving the quality of retrieval.
- D. Incorrect.
Increasing the temperature parameter affects the diversity of the responses but does not address the issue of noisy or irrelevant content in the source documents.
- E. Correct.
Using embeddings to rank content based on semantic similarity ensures that only the most relevant information is prioritized for retrieval, improving the overall response quality.