电信网络诈骗预防线索分析数据

当前电信网络诈骗存在的覆盖不全面、不精准、触达不及时、沟通不便捷、数据不闭环，进而导致反诈宣防和劝阻拦截工作效果无法科学评估和预判风险的问题。通过大模型和流式计算引擎进行异常模式识别，结合业务场景构建涉诈话术特征库，最终形成动态更新的诈骗风险预警数据及模型，可实现精准分析反诈数据，并指导反诈宣传和劝阻工作。1.数据采集：从百应客户互动平台系统数据库中，提取反映每次触达及互动过程的数据，包括触达次数，成功次数，预警准确数等；
2.数据处理：将采集到的数据，去除异常值和错误值，根据模块进行分类，并执行计算。
多模态触达发起数 = sum(通过安盾系统发起的触达次数)
多模态触达成功数 = sum(系统发起，并成功被接通的次数)
多模态触达成功率 = 多模态触达成功数/多模态触达发起数
预警准确数 = sum(触达成功后，市民在过程中表示触达信息准确的肯定次数)
预警准确率 = 预警准确数/多模态触达成功数
被骗数 = sum(触达成功后，市民表述自己被骗的次数)
被骗率 = 被骗数/多模态触达成功数
宣传成功数 = sum(触达成功，且触达内容全部播放完成的次数)
宣传成功率 = 宣传成功数/多模态触达成功数
3.数据应用：根据触达成功率、预警准确率等指标，判断当前触达和预警系统的质量，根据被骗数和被骗率指标，识别高危人群，指导劝阻工作。

Current telecom fraud suffers from multiple critical shortcomings, including incomplete coverage, imprecise targeting, delayed outreach, inconvenient communication, and non-closed-loop data processing. These issues further impede the scientific evaluation of the effectiveness of anti-fraud publicity, prevention, and intervention work, as well as the accurate prediction of related risks.

To address these challenges, this dataset utilizes large language models (LLMs) and streaming computing engines to identify abnormal patterns, and builds a feature library of fraudulent scripts in combination with actual business scenarios. Ultimately, dynamically updated fraud risk early warning data and models are generated, enabling precise analysis of anti-fraud data and providing actionable guidance for anti-fraud publicity and intervention efforts.

1. Data Collection: Data reflecting each outreach and interaction process is extracted from the database of the Baiying Customer Interaction Platform system, including the number of outreach attempts, successful outreach attempts, accurate early warning cases, and other relevant metrics.

2. Data Processing: Outliers and erroneous values are first removed from the collected data, which are then categorized by modules and subjected to quantitative calculation. The specific indicator formulas are listed below:
- Multimodal outreach initiation count = Sum of outreach times initiated via the Andun System
- Multimodal outreach success count = Sum of successfully connected system-initiated outreach attempts
- Multimodal outreach success rate = Multimodal outreach success count / Multimodal outreach initiation count
- Accurate early warning count = Sum of affirmative responses from citizens confirming the accuracy of the outreach information after successful outreach
- Early warning accuracy rate = Accurate early warning count / Multimodal outreach success count
- Fraud victim count = Sum of times that citizens reported having been defrauded after successful outreach
- Fraud victimization rate = Fraud victim count / Multimodal outreach success count
- Successful publicity count = Sum of successful outreach attempts with the entire outreach content fully played
- Publicity success rate = Successful publicity count / Multimodal outreach success count

3. Data Application: The quality of the current outreach and early warning system is assessed using indicators such as the outreach success rate and early warning accuracy rate. High-risk groups are identified and targeted intervention work is guided based on the fraud victim count and fraud victimization rate.

搜集汇总

背景与挑战

背景概述

该数据集由浙江百应科技有限公司提供，包含632条记录，每日更新，主要用于电信网络诈骗的预防和劝阻工作。数据通过大模型和流式计算引擎进行异常模式识别，构建涉诈话术特征库，形成动态更新的诈骗风险预警数据及模型。