Docs

DataAgent (OnPremise)

Unstructured Data Labeling

Overview of Unstructured Data Labeling

Unstructured data labeling, as a business empowerment tool supporting intelligent processing of multimodal data, helps enterprises integrate unstructured data from multiple channels and convert text, audio, image, video, and other unstructured data into standardized business labels. Supports multi-source Data Integration, Custom Labeling System, and labeled Data Output, enabling enterprises to build reusable label assets and drive intelligent analysis and decision-making.

Function introduction

By integrating unstructured data from multiple channels such as enterprise WeChat conversations, outbound call recordings, and intelligent agent conversations, the tool provides conversation overview, original conversation query, conversation labeling, insights into user evaluations and suggestions regarding products, services, and marketing, as well as conversation quality assessment to inform business operations; it also uncovers individual and group user intentions and preferences, deepens user understanding, improves user operations, and supports enterprise growth. Supports labeling for Article, Image (such as product images), Video (such as product introduction videos), and other unstructured data.

Application Scenario

Widely applicable across multiple industries and scenarios. In the customer service domain, label extraction from large volumes of conversation data enables rapid understanding of customer needs and issue types; in the marketing domain, analysis of social media data and user comments allows extraction of user interest labels to support targeted marketing.

Prerequisites

When using the unstructured data labeling feature of Conversation Assistant, you must also activate VeCDP and BytePlus Ark products, as the unstructured data labeling function relies on these two products. For specific activation methods, please refer to Customer Data Platform Product Online Purchase Guide and BytePlus Ark Large Model Service Quick Start.

Dependent Products	Feature Dependency Description	Notes
VeCDP	The unstructured data labeling feature is implemented based on VeCDP's Visual modeling functionality. By creating Visual modeling tasks, you can invoke restful operators for data processing and large model operators for intelligent Label output. Unstructured data must be integrated into the VeCDP Dataset, which consolidates the unstructured data to be labeled. The Label results from unstructured data labeling can be stored in VeCDP as Label data, facilitating further usage.	VeCDP and the unstructured data labeling feature must be under the same account.
BytePlus Ark Platform	The unstructured data labeling feature supports integration with BytePlus Ark Platform via large model operators. Users only need to complete the large model access point and Token configuration on the platform to enable intelligent labeling capabilities and generate Label results in real time.	The currently supported large model types are those already available on your Ark Platform instance. Third-party large models must comply with the OpenAPI standard protocol.
Other Products	Currently, direct import of unstructured data into VeCDP as a Dataset is not supported. You need to first store unstructured data in database tables, and then create a VeCDP Dataset by connecting to these tables. Therefore, you can select an appropriate database table storage product based on the methods currently supported by VeCDP for creating Datasets.

Glossary

Noun	Explanation
project	A Project is an independent "space" for using the product. Except for sharing server hardware resources, all other resources are independently isolated between Projects. For example, user permissions and data resources in Project A and Project B are different.
Project Center - User	The product user needs to be created in BytePlus Access Control and includes information such as Project, account, email, and phone number.
Project Center - User Group	A User Group refers to a group of users with special business significance. It belongs to a specific Project and cannot be used across Projects. It needs to be independently created in Project Center Authorization Management.
Module	The product's Functional Modules correspond to their respective menus. Module usage permissions can be authorized in Project Center.
Resources	Resources produced by users in the product include Label, Dataset, Segmentation, insight report, and Visual modeling tasks. Resource permissions can be authorized in Project Center (Segmentation and user insight Authorization are performed on the details page).
Data	Specifically refers to Label, Dataset, etc. in the product. Rules can restrict the range of ID assets users can access; you can also control the data application scenarios, such as which tables and columns users can access.

Operational Workflow

The main operational workflow for configuring unstructured data labeling tasks is as follows:

Enable the VeCDP Customer Data Platform. For detailed instructions, see Enable VeCDP and Authorization Configuration.
Before using the unstructured data labeling feature, you must first create the unstructured data to be labeled as a VeCDP Dataset. For detailed instructions, see Data Preparation.
Create a Visual modeling task and configure the Data Connection. For detailed instructions, see Task Creation.
Configure the relevant operators to process unstructured data. For detailed instructions, see Unstructured Data Processing Operators Configuration.
Configure the final output and storage format for unstructured data. For detailed instructions, see Processing Result Output Configuration.
For permission operations related to account, user, and Label features in unstructured data labeling, see Authorization Configuration.

Last updated: 2026.06.30 11:48:49