Draft article not currently submitted for review. This is a draft Articles for creation (AfC) submission. It is not currently pending review. While there are no deadlines, abandoned drafts may be deleted after six months. To edit or make changes to this draft, simply click on the "Edit" tab at the top of the window. To be accepted, a draft should: Show the subject qualifies for a Wikipedia article by using multiple sources that meet four criteria. The sources should be (1) reliable (2) secondary (3) independent of the subject (4) talk about the subject in some depth. For some topics, there are alternative criteria. Be written from a neutral point of view Respect copyright and do not plagiarize. Do not copy-paste. It is strongly discouraged to write about either yourself or your business or employer. If you do so, you must declare it. Where to get help If you need help editing or submitting your draft, please ask us a question at the AfC Help Desk or get live help from experienced editors. These venues are only for help with editing and the submission process, not to get reviews. If you need feedback on your draft, or if the review is taking a lot of time, you can try asking for help on the talk page of a relevant WikiProject. Some WikiProjects are more active than others so a speedy reply is not guaranteed. How to improve a draft Wikipedia:Contributing to Wikipedia – a basic overview on how to edit Wikipedia. Help:Wikitext – how to use the markup Help:Referencing for beginners – how to include references Wikipedia:Article development – how to develop your article Wikipedia:Writing better articles – how to improve your article Wikipedia:Verifiability – make sure your article includes reliable third-party sources You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Add tags to your draft Editor resources Easy tools: Citation bot (help) | Advanced: Fix bare URLs Last edited by Qwerfjkl (bot) (talk | contribs) 53 days ago. (Update) Submit the draft for review!

RAG-Fusion is a technique in natural language processing that extends retrieval-augmented generation (RAG) by combining it with reciprocal rank fusion (RRF) to improve the relevance and comprehensiveness of responses generated by large language models (LLMs).

RAG-Fusion was introduced in a 2024 paper by Zackary Rackauckas, an AI and business development research intern at Infineon Technologies.^[1] The technique was developed to address a limitation of standard RAG: a single user query often fails to capture the full scope of information needed to answer complex, knowledge-intensive questions.^[1]

Rather than retrieving documents based on a single query, RAG-Fusion generates multiple variations of the original query using an LLM, retrieves documents for each variation, and then combines the resulting ranked lists using reciprocal rank fusion before passing the fused results to the generative model.^[1]

The RAG-Fusion pipeline consists of three main stages:^[1]

1. Query generation — An LLM generates multiple variations of the user's original query, intended to capture different perspectives on the same information need.
2. Parallel retrieval — Each generated query is used to independently retrieve a ranked list of relevant documents from a vector database or other retrieval system.
3. Reciprocal rank fusion — The multiple ranked lists of documents are merged into a single ranking using the reciprocal rank fusion formula, which sums the inverse of each document's rank across the different result lists.

The fused, re-ranked list of documents is then provided to the LLM as context for generating the final response.

In the original study, RAG-Fusion was evaluated by manually assessing generated answers for accuracy, relevance, and comprehensiveness.^[1] The author found that the technique excelled in accuracy and comprehensiveness, but occasionally strayed in relevance when the LLM generated queries that were not closely related to the original question.^[1]

A follow-up 2024 study by Rackauckas and colleagues proposed an automated evaluation framework called RAGElo, which uses an Elo-based ranking system and LLM-as-a-judge methods to compare RAG-Fusion against standard RAG.^[2] This study found that RAG-Fusion significantly outperformed standard RAG in answer completeness, based on both expert annotations and automated Elo-based scoring, but underperformed standard RAG in precision.^[2]

RAG-Fusion has been applied in enterprise chatbot settings, including a deployment at Infineon Technologies intended to help engineers, account managers, and customers obtain product information more efficiently.^[1] The original author proposed extending such systems to support additional languages and to better incorporate multimodal technical documents, such as PDF datasheets, as retrievable text.^[1]

As with standard RAG, RAG-Fusion does not eliminate the risk of hallucination in LLM-generated responses. Generating multiple query variations also increases computational cost compared to single-query retrieval, since each variation requires a separate retrieval pass. Evaluation has also shown a trade-off between completeness and precision: while RAG-Fusion tends to produce more complete answers than standard RAG, it can underperform on precision.^[2]

• Retrieval-augmented generation
• Information retrieval
• Large language model

16