To create a chatbot powered by generative AI, the knowledge base (Knowlegde Base) is a key element…
Customer relations teams’ expectations of the chatbot :
1/ answer all customer questions
2/ make relevant answers
3/ be immune to hallucinations
Did you know that all this can be configured in the knowledge base? 🙂
Here’s some feedback from our customer relationship automation projects. Allez, go!
❓What is a knowledge base?
Simply put, a knowledge base is a library of practical information about a specific product, service or topic relevant to your business.
For a chatbot, the knowledge base is seen as the source of truth it will use to respond.
Discover our solution Knowledge base
🤔 Have you seen it before?
The data in your knowledge base can come from a variety of sources: website, FAQ, CRM, working documents…
This data comes in a variety of formats: text, images, tables, URLs, videos…
🚩 So, what’s the problem?
The secret of a well-functioning LLM chatbot is the quality of its knowledge base.
What to do?
➡️ For the LLM to be able to “think” from the data in the knowledge base, the data must be structured and prioritized. In other words, they must be cleared of unnecessary information and ordered according to importance. The challenge here is to 1/ collect the data, 2/ clean and structure it, and 3/ merge the various documents into a coherent whole.
➡️ Data must correspond to the questions asked by users. There’s no point in collecting, cleansing and structuring data if it doesn’t provide relevant content for user responses. This means having a clear idea of the customer’s questions, and avoiding off-topic information that can pollute the knowledge base and lead to wrong answers.
➡️ Terminology and technical jargon specific to the company’s activity must be defined in the knowledge base so that the LLM can reason without ambiguity. In the health insurance sector, for example, the term “TP” must be precisely defined because the intended meaning is “Tiers-Payant”, otherwise the LLM may choose to give it the meaning of “Travaux Publics” or “Taxe Professionnelle”.
Our #advice for your chatbot development project powered by generative AI :
👉 Analyze the history of questions asked by your users to get a precise list of customer needs 🔎
👉 From this analysis, identify the documents and content to be placed in the knowledge base 🎯
👉 Remove unnecessary information from documents, then transform each document into a logical, hierarchical semantic tree 🪚
👉 Place this content in the knowledge base and segment it into several parts according to the processes to be automated via the chatbot ⚙️
👉 Regularly update your knowledge base 🔄
Are you about to launch a generative AI development project?
Talk with us about your project ⬇️
