I have a chatbot feed in Knowledgebase from pages of a website. ( It is a website chatbot assistent). I have uploaded all pages of the website in the knowledgebase.
The assistent reply correctly to questions asked, in the matter of content.
However, is instructed to display the source link of the page ( means the product page of an woocmmerce site). - BUT this task is not managed accurately. The Piackaxe invent URL which not exist on my website. How can i manage to solve this !
@doreloprisan the problem could be with the prompt or the way the knowledge base has been created (chunking).
To make sure the pickaxe retrieves the URL correctly, my suggestion would be to create an Excel document with 3 columns: description of the link, URL, and content (do not add empty rows).
If you then upload the file, each row in your excel is added as a chunk. The URLs are now separate and it is easier to retrieve them.
Thank you for your reply. However, the site could have lot of pages. I have added my pages of the website through scraping system of Pickaxe (Upload webpage). That’s why I do not understand why the source URL is not displayed correctly. Any idea how to craft the Prompt to avoid inventing URL -s which not exist ?
@doreloprisan unfortunately, the scraping doesn’t necessarily work all the time. It depends on how the target website is structured. Without looking at the pickaxe, it is hard to pinpoint the problem.
Maybe you can start by adding in the prompt something like:
The information you provide should only come from the www.example.com website. If you provide users with any URL, make sure they are related to this website (for example, www.example.com/anotherexample).
If available, you can also add a sitemap of the website as a reference
Hi, thanks for reply.
When I tried to add sitemap, error message: The filetype xml is not accepted to be uploaded in knowledgebase.
When I insert in prompt something like: check URL to be sure is related to website, I have a message from chatbot like: I don’t have acces to the internet …