Hi everyone.

You can check the nodes it is fetching based on your query and then check further from there:

Thank @WhiteFang_Jr .

However, one issue is that I found the query_engine.query does not return the correct node with similar keywords in the docx file.

Not all, but there are a few places where I cannot query out.

index = load_index_from_storage(storage_context)
query_engine = index.as_query_engine(text_qa_template=qa_prompt_tmpl, streaming=True, service_context=service_context)

response = query_engine.query(f"{input_text}")
print(response.source_nodes)

Okay so you will have to go through the nodes and check how you can improve your data for for more better data.

You could check this: https://gradient.ai/blog/rag-101-for-enterprise on how to improve your RAG experience

Thank you. I will explore those methods further. But is there any similar section in the llamaindex documentation?

Not in the documentation But i think there was a blog in LlamaIndex blog. Let me check

https://blog.llamaindex.ai/evaluating-the-ideal-chunk-size-for-a-rag-system-using-llamaindex-6207e5d3fec5

This one

I am using this piece of code with a modified chunk_size, and I'm encountering a situation that I can't find an explanation for:

query_engine = RetrieverQueryEngine.from_args(retriever, text_qa_template=prompt_tmpl, service_context=service_context)
response = query_engine.query(input_text)

In my data, it talks about a musician named A. It also mentions that 'Musician A once acted in movie B'. However, when I query 'Has A ever acted in any movies?', I receive the answer 'There is no data mentioning that Mr. A has ever acted in movies'. But when I ask 'Has he ever acted in any movies?', I get back the result 'Mr. A has acted in a movie called B'. I am very curious why there are two different results like this.

Check the nodes which are returned in the resposne and verify if correct nodes are picked for all these scenario or not.

You can print nodes like this: print(response.source_nodes)

I am taking 6 nodes, then there are 2 different nodes with the 2 questions above.

retriever = QueryFusionRetriever(
indexes,
similarity_top_k=6,
num_queries=3,
mode="reciprocal_rerank",
use_async=True,
verbose=True,
)

///////////////////////////////////////////
There is one more issue. When I describe at length, the result cannot capture the entire description but only gets the beginning. How can I make it capture everything