Knowledge Base
By importing various knowledge documents into the bot and training it, the bot's ability to answer questions can be effectively improved.
It supports copying documents that are not available from other bots in the same organization to the current bot to become the current bot's knowledge documents. This process does not consume credits.
Documents
Current documents support importing documentfile, excel file, URL, text and Q&A in the form of documents into the bot for training.
Go to the [Bot - Knowledge - Documents] page and [Select Document Type - Add Dataset] to add corresponding type of knowledge documents.
After a document is successfully added and trained, its status will become "Enabled".
Document Files
If you have existing document files, you can import the files into the bot as knowledge. Supported formats: pdf, doc, docx, txt, md.
Add Files
Click the upload area or directly drag and drop the file into the upload area and click "Next".
Slice rule
Custom slicing of document content is supported. You can choose to slice by "Maximum Token Number" or "Identifier".
Slice by Maximum Token Number: Define the maximum number of tokens for a single slice, and the system will slice by this token number. The image slices by 600 tokens, and the corresponding token number is displayed on the far right.
Slice by Identifier: Define slice identifiers, and the system will slice by identifiers. If a slice exceeds the maximum token length defined by the system, it will be automatically truncated. The image above slices by identifier (======). After selecting the identifier, you need to click "Update Preview" to preview according to the currently selected rules.
Click 【Save】 to upload the file for training.
Edit Documents
In the operation column of the knowledge document, select [More - Document Edit] to edit the document content online.
After completing the online edit [Save] to retrain the document content.
Excel File
If you have an existing spreadsheet, you can import the file into the bot as knowledge. Supported formats: csv, xls, xlsx.
Note:
The system only reads the first worksheet in the file.
Be sure to ensure the table starts with cell A1, otherwise the data slice results may not meet expectations.
Click or drag the file to this area to upload, and click "Next".
Slice by row as header, then slice by row, each slice will contain header information and row information. If a slice exceeds the maximum Token length defined by the system, it will be automatically truncated. By default, "Row 1" is the header, and only consecutive multiple rows starting from "Row 1" can be selected as the header. Click "Update Preview" to preview the slicing.
Click 【Save】 to upload the file for training.
URL
If your knowledge exists in the form of a website, you can enter the URL and the system will crawl the information on the web page and import it to train the bot's knowledge.
Please note that in some special cases, web content cannot be crawled, including but not limited to:
Login required to access
Anti-crawler mechanisms
......
Therefore, we do not encourage you to make web pages your top priority for knowledge management.
Add URL
There are two ways to add website information as knowledge.
In 【Whole Site Import】, you only need to enter a domain name, click 【Identify URL】, and the system will identify all subpages under the domain name. You can delete webpages that you do not need to import into the bot from the list below.
Select the slicing method (the operation is the same as for document files), and click 【Update Preview-Save】 to import the content of all webpages in the list and train them as bot knowledge.
In【Batch Import】, you can only import certain specified links as bot knowledge. The submission operation is the same as 【Whole Site Import】.
Edit Webpage Content
Same as 【Files】.
Refresh
In the operation column of the knowledge document, click [More - Refresh] to refresh the knowledge for this webpage type.
This feature is unique to the webpage type of document. Because website content may change, this feature allows you to retrieve the latest content of the webpage again and train it as bot knowledge.
Text
Add Text
You can directly enter the knowledge information online in text form.
Select the slicing method (the operation is the same as for document files), and click【Update Preview-Save】to import the text content and train it as bot knowledge.
Edit Text
Same as [Document Files].
Q&A
Q&A refers to saving and managing knowledge content in a "question and answer" format.
Compared with files, text or websites, the advantage of Q&A is that the knowledge data is structured, which means it is clearer to manage, has lower training costs, and can return more accurate answer content in specific questioning scenarios. The disadvantage is that because it is structured data, it is more rigid and less flexible.
If your knowledge information contains a lot of question and answer type content (such as FAQ), you can consider using this type of knowledge document first.
Add Q&A
Supports document import, template import and document to Q&A conversion. New Q&A pairs can be added to existing documents.
Select [Document Import] and enter multiple Q&A pairs. You can preview the slicing on the right after adding them.
Organize the Q&A file according to the template format, select [Template Import] to upload the template file. You can preview the slicing on the right after successful upload.
Select [Document to Q&A] to upload the document, and the system will parse the document content and convert it into multiple Q&A pairs using LLM capabilities. This feature will consume credits.
Click [Save] to import the content into the bot for training.
Edit Q&A
Same as [Files], but the edited format maintains the "Q" and "A" structure, consistent with when adding.
Vector
In vector management, you can view the segmentation of each knowledge document, and set the vector search setting enabled or disabled for the specified segments or the entire knowledge. For closed knowledge or fragments, Bot will not retrieve and return the knowledge of this part.
In the operation bar of the knowledge document, select [More - Vector].
Here, vector search can be turned on or off for segments and knowledge.
Vector Search
[Vector Search] is a module for you to test the effect of knowledge vector search. You can execute a vector search by entering information. The system will return the search result segments for you according to the vector similarity score from high to low.
Last updated
