Social Fraud Detection LLM Solution: Features and Parameters-Fraud Detection(Fraud Detection)-阿里云帮助中心

The Social Fraud Detection risk identification solution uses Large Language Models (LLMs) to understand and analyze complex semantics. It precisely detects hidden risks in chat contexts—such as fake identities and fraudulent traffic diversion—and quickly identifies users involved in fraud during social chats. Your business can act on the service’s response.

Service Event Parameters

Service event parameters are the request parameters in the Service and ServiceParameters fields of common parameters. The Social Fraud Detection LLM Solution uses the following Service value: llm_fraud_detection. ServiceParameters is a JSON object. Pass the following request parameters.

Important

For maximum compatibility with various input formats, Alibaba Cloud Fraud Detection does not validate the business format of string inputs. You must validate the business format yourself. For example, for the mobile field, your application must verify that the input matches the Chinese mainland mobile phone number format—11 digits, starting with 1.

Field Name	Description	Data Format	Example	Required
dialogs	Context of private chat messages sent by the user. Maximum length is 2000 characters (counting only the content field). Split longer content into multiple requests.	String	Concatenate user role and message content. Example: `[ { "role": "A", "content": "Hello" }, { "role": "B", "content": "Hi, where are you from?" }, { "role": "B", "content": "I am xxx" } ]`	Yes
accountIdA	Account ID for user A. Used only for association.	String	You can pass a unique external identifier, such as a mobile phone number.	No
accountIdB	Account ID for user B. Used only for association.	String	You can pass a unique external identifier, such as a mobile phone number.	No
mobile	Default: mobile phone number in the Chinese mainland. For numbers outside the Chinese mainland, include the country code in `country code-mobile number` format.	String	Chinese mainland mobile number: 138**1111 Mobile number outside the Chinese mainland: 001-718123**	No
mobileMd5	MD5 hash of an 11-digit mobile number starting with 1. Result is a 32-character lowercase string of letters and digits.	String	Before encryption: 180**0000 After encryption: e7beea81b7a03b38508428fbeeb3**	No
deviceToken	Device token obtained using the Device Risk SDK.	String	MzQvo1d7scyZ3tl_RcJZo_QOytAjy1LWRRLoRKo5oZSoo_JGj1ZoR5JGoRo5jcdn57gV5kxVRcLER5RQoZSvRZZQRcROjcMW5csZR_RGy_55RKJ_oooqZ7dSV5gRnKxOV7eWVQQjRtlRQoAjRcM0	No
ip	Public IPv4 address of the client (user side) at the time of the event.	String	42.120.XX.XX	No
email	Email address.	String	admin****@aliyun.com	No
operateTime	Operation timestamp in seconds, in Greenwich Mean Time (GMT). Note If scanning historical data, set this field to the historical operation time. This avoids false positives caused by incorrect timestamps.	Long	For example, GMT 2018-04-01 12:00:00 corresponds to timestamp 1522555200.	No

Response Parameters

The Fraud Detection service returns risk tags, risk reasons, and risk IDs. The business meaning of the Data field’s tags and extend values is explained in the action recommendations below. These recommendations reflect best practices from the Alibaba Cloud risk control team.

Note

You can choose actions flexibly based on your business needs.

Field Name	Description	Values
tags	Risk tag	Example: `is_fraudPersona`, and others
extend:riskRole	Risk account. Role name of the risky user in the `dialogs` field.	Example: `A`, and others
extend:reason	Risk reason. An LLM-generated description of the detected risk. Usually no more than 500 characters.	Example: "In this conversation, B claims to work in securities investment—a high-risk profession commonly seen in social fraud scenarios. Also, B mentions…"

The possible values for the Data field’s tags field appear in the table below.

Tag Value	Tag descriptions
is_fraudPersona	Fake identity. Includes: Trusted roles, such as military personnel or civil servants. Wealthy roles, such as second-generation heirs or first-generation entrepreneurs. High-risk professions, such as internet O&M engineers or investment managers.
is_fraudDiversion	Fraudulent traffic diversion. Includes evasive tactics like diverting users to external sites or third-party niche apps.
is_sexual	Sex-related solicitation. Includes risks such as sex-for-money transactions and nude-chat scams.

Note

If the tags field in the response is empty, no risk was detected for this request.

For more information about common response parameters, see Common Response Parameters.