# Introduction to Document Datasets If you would like your conversational agent to assist users by leveraging your existing document knowledge — providing industry or platform\-specific information in the responses — you can use Document Datasets to import unstructured data and link them to your application. Unlike Item Datasets which store structured data, Document Datasets support the storage and processing of **unstructured data** (document content). The following formats are currently supported: * **formatted document data:** files with pdf, doc, docx, ppt, pptx extensions * **plain text document data:** files with md, html, txt extensions * **spreadsheet files:** files with xlsx, xls, csv, jsonl extensions * **Image files:** bitmap files with extensions JPEG, PNG, WEBP, BMP, TIFF, ICO, DIB, ICNS, SGI, JPEG2000 You can empower the following scenarios with document knowledge: * **E\-commerce Shopping Guide:** Import membership policies, after\-sales policies, product details, and promotion documents to help the Agent answer common questions about member benefits, return processes, and product specifications. * **Vertical Content Search:** Import industry\-specific knowledge to help the Agent understand and answer professional questions in a specific field, leading to better query interpretation and search planning. :::tip \*difference between spreadsheet files and item datasets While both are "structured data" with defined and fixed column formats, they serve different purposes in Viking AI Search: * **Item Datasets:** Used to build structured index data for **Search and Recommendation**. Fields can be designated for specific uses (semantic retrieval, precise filtering, user behavior preference analysis, etc.). If you want your data to be **candidate content** for search or recommendations, you **must** use an Item Dataset. * **Spreadsheet Documents:** Act as a **Knowledge Base for Q&A**. Suitable for uploading Q&A pairs, reviews, FAQs, or product manuals. These serve as a knowledge source for answering user questions in dialogue but **cannot** be used as search or recommendation candidates. ::: # Create a Document Dataset You can open the Document Dataset tab under Data Configuration in the application interface to create a document dataset and directly link it to the application, or create a document dataset under Application \> Dataset \> Document Dataset tab and manually link it to an application.

Create in Application

![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/cad83320b588498a820dda33b85bf60a =5090x)

Create in Dataset

After clicking create, you need to name the document dataset and import a batch of document files. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/8f595b805a06489d8fe3b404b29b8639 =4744x) The system will perform format and size validation during import. See [文档格式和大小限制](/docs/85296/1894022#ac881958) for supported document formats and size limits. You can also enable the document import feature on the Document Dataset detail page under Datasets. # Document Processing and Storage Once imported, the system automatically performs intelligent processing, including reading, parsing, slicing, index construction, and vector storage to ensure the content is searchable. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/70fe9ed2e093446b93c94decd28594e7 =5096x) Original documents will be processed into markdown format. Once processing is complete, you can click the `View` button to inspect the processed text content. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/67fa6f8afb044f7587b85bb322e5beef =5098x) ## Delete Documents You can delete documents in both the application and dataset document interfaces. After clicking the `Delete` button, you will be asked to confirm the sensitive operation. Click `Confirm` to delete the document. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/a5543611218049e3bd5780fd9bbd8017 =5078x) # Document Format and Size Limits Document data features are currently available for free during the Beta phase. We support uploading and processing the following document types: |**Document Type** |**Size Limit** | |---|---| |**pdf** |A single document must not exceed 100MB or 1000 pages | |**doc** |A single document must not exceed 100MB or 1000 pages | |**docx** |A single document must not exceed 100MB or 1000 pages | |**ppt** |A single document must not exceed 100MB or 1000 pages | |**pptx** |A single document must not exceed 100MB or 1000 pages | |**markdown** |Single document no larger than 5 MB | |**txt** |Single document no larger than 5 MB | |**html** |Single document no larger than 5 MB | |**xlsx** |No more than 50 sheets, total rows no more than 30,000 | |**xls** |No more than 50 sheets, total rows no more than 30,000 | |**jsonl** |30,000 lines, no line exceeding 65,535 characters | |**Image files (JPEG, PNG, WEBP, BMP, TIFF, ICO, DIB, ICNS, SGI, JPEG2000)** |Single file size no larger than 8MB|\ | ||\ | |* Aspect ratio (width/height): limited to the range [1/100, 100]|\ | |* Side length limited to the range [10, 6000] pixels |