RAGFlow retreival process doubt

[harsh788]

I have pushed in almost 8000 chunks to a document in a database through http api. In the api call, I did not pass any keywords with any of the chunks. So, if I don't have any keywords, what is the significance of vector_similarity_weight variable during retrieval thorugh the api?

[dosubot[bot]]

If your chunks don't have keywords, RAGFlow's retrieval will rely entirely on vector similarity. The vector_similarity_weight parameter then becomes the main factor controlling how much influence vector similarity has in ranking and returning results—since there are no keywords, it's effectively the sole scoring method used for retrieval in your case. You can use this parameter to fine-tune how strict or broad the vector-based matching is for your queries. More details here and here.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[harsh788]

Okk, so in case of no keywords, the retrieval depends entirely on vector_similarity_weight. But, if it is a weighted average of the two values and there are no keywords, why am I getting different results for different values of vector_similarity_weight?

[dosubot[bot]]

When there are no keywords, RAGFlow relies entirely on vector similarity for retrieval. The vector_similarity_weight parameter still affects the results because it directly scales the normalized vector similarity scores used for ranking. Even though the weighted average formula expects both keyword and vector scores, if keywords are missing, only the vector part is used—so changing vector_similarity_weight changes how strictly or loosely the system ranks and filters results based on vector similarity alone. This means different values can lead to different chunks being included or excluded, or change their order in the results, even without keywords present. You can see more about this behavior in the official discussions and issues: discussion, issue, and API details.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[eyyyting]

Hi @harsh788
When performing retrieval, the system automatically extracts key terms from the user’s query . By default, the system handles keyword extraction and matching for you. Of course, you can also manually add keywords to further improve the accuracy of the results.

[harsh788]

Hi @eyyyting

the system automatically extracts key terms from the user’s query

But what is the use of it if I have not supplied any keywords while pushing chunks to data to my knowledgebase? Against what will the keyword similarity search be performed?