What is RAG?
The term "RAG (Retrieval-Augmented Generation)" has been heard frequently lately.
Simply put, it is a mechanism where generative AI, when generating answers, searches for necessary information from pre-prepared external sources (such as internal manuals or knowledge bases) and creates answers based on that information.
By incorporating external information, it becomes possible to generate more reliable answers than when using generative AI alone.
Regular generative AI answers user questions using only the data it has previously learned. Therefore, it may provide outdated information or respond with inaccuracies about topics it has not been trained on (hallucinations).
On the other hand, RAG searches for relevant information in real time after receiving a question and answers based on that content. Therefore, users can obtain more accurate and up-to-date information.
1-1. Uses of RAG
So, in what situations is RAG used?
● Customer Support
When RAG is used in chatbots for corporate customer support, it can search for necessary information from internal manuals and FAQs to provide answers, enabling more accurate responses.
● Internal Operations
Information can be extracted from manuals for employees and only the necessary information can be provided in real time.
For example, by using a chatbot that leverages RAG (commonly called a "RAG Chatbot"), when a new employee asks, "Please tell me the procedure for this task," it searches the relevant information from internal documents and provides an answer.
● Marketing and Market Research
It can search for and summarize market trends and movements from web articles and news, conveying them clearly to marketing personnel.
In this way, RAG is utilized in various tasks, gathering information and enabling generative AI to provide intelligent answers, making it a noteworthy mechanism.
Essential Efforts to Improve Accuracy for Effective RAG Utilization
RAG is a convenient system, but it is very important not only to introduce the technology but also to simultaneously work on improving its accuracy.
This is because the quality, structure, and processing methods of the information targeted by RAG's search can affect whether appropriate answers can be provided to users.
For example, if the information found in external sources is actually unrelated to the user's question, the generated answer will be off the mark.
Also, if only a portion of the information is extracted to generate an answer, there is a risk that the original meaning will be conveyed incorrectly. Such unclear answers can cause user confusion and lead to risks such as operational judgment errors or delays in work.
Furthermore, repeated incorrect answers can undermine the reliability of the RAG chatbot itself, potentially leading to a situation where it is ultimately no longer used.
Therefore, to operate an RAG chatbot at a practical level, it is essential to work on improving accuracy so that accurate and reliable information can be provided consistently.
Methods to Improve RAG Accuracy
To maximize the strengths of RAG, it is necessary to organize the input information and improve the search and generation processes.
Here, we introduce effective methods to enhance the accuracy of RAG.
● Organize the input documents
RAG processes the searched documents so that the generative AI can understand and utilize them, but if the original documents are miscellaneous or the expressions are ambiguous, the information cannot be used correctly.
Therefore, it is important to organize internal documents and manuals so that their meaning is clear to anyone who reads them.
・Make the subject and predicate clear
・Use bullet points to organize information
・Avoid ambiguous expressions (such as "this" or "that")
・Clearly distinguish types of information such as procedures, cautions, and supplementary notes
● Choose the optimal search method
・Keyword search: Find documents containing specific words
・Vector search: Find documents based on semantic similarity
・Hybrid search: Combine both methods to improve accuracy
● Chunk the documents
By dividing documents into fixed sizes, it becomes easier to search for more appropriate information.
● Preprocess data such as diagrams and tables
・Tables are written using Markdown table syntax
・Figures are converted into text as descriptive captions
・Flows are expressed using bulleted lists
● Add information to improve searchability
・Document category
・Update date
・Related keywords
・Target users
● Continuously improve
・Analyze causes of incorrect answers
・Add and update documents
・Review search structure
・Improve prompts
Creating documents that are easy to understand for both generative AI and humans—"documents friendly to both"—is important.
A case study on document organization with an eye toward building RAG
● Issues and Background
・Manuals are difficult to understand and formats are inconsistent
・Want to utilize generative AI chatbots in the future
● Initiatives
・Creation of model manuals
・Establishment of manual creation guidelines
● Future Outlook
We plan to roll out the organized manuals company-wide and load them into the RAG chatbot for verification.
For consultations on manual organization, contact Human Science
Human Science provides one-stop support from Japanese manual creation to English and multilingual translation.
・Want to improve existing manuals
・Want to create English manuals step by step
・Want to expand Japanese manuals into multiple languages
Feature ①: Extensive Manual Production Experience
NTT DOCOMO, INC.
Yahoo Japan Corporation
Yamaha Corporation
Manual Production Case Studies｜Human Science
Feature ②: Support from Research and Analysis to Output
Experienced consultants create the optimal manual starting from the hearing process.
Manual Evaluation, Analysis, and Improvement Proposal Service
Feature ③: Support Including Adoption Assistance
We support adoption through updates and seminars even after the manual is created.
Thank you for reading until the end.
I hope this blog serves as a helpful guide for creating easy-to-understand manuals.
