基于通信维度的信用风险评分数据

羽乐科技基于用户授权的情况下，获得用户匿名陌电服务场景的数据，将相关信息集合形成匿名化处理后的分析/统计类信息，加工形成应用于金融服务场景的基于通信维度的信用风险评分数据产品。金融机构基于其客户授权的情况下，调用羽乐科技上述基于通信维度的信用风险评分数据产品，用于辅助判断客户信用风险，帮助金融机构解决识别客户信用风险问题，提升金融机构风控水平。计算近35天陌电查询次数、近30天命中泛金融标签(偏好客群)占比、近30天命中泛金融标签(偏坏客群)占比、近30天陌电查询PV均值、近30天陌电查询标准差、近30天命中泛金融标签次数、近30天命中疑似风险标签中位数、近30天陌电查询PV均值、近30天陌电查询标准差等统计量，再基于多个统计量，利用算法模型加工为分值评分，用于度量疑似信用风险程度的概率。标准分=a*近35天陌电查询次数+b*近30天命中泛金融标签(偏好客群)占比-c*近30天命中泛金融标签(偏坏客群)占比+（d*近30天陌电查询PV均值*近30天陌电查询标准差）-（e*近30天命中泛金融标签次数*近30天命中疑似风险标签中位数）+（f*近30天陌电查询PV均值*近30天陌电查询标准差）其中，a、b、c、d、e、f为各个指标的权重，可以根据实际情况进行调整。最终计算分值在[0,320)区间，则代表用户信用评分极好，风险极低；[320,560)区间，则代表用户信用评分较好，风险较低；[560,840)，则代表用户信用评分一般，有一定的风险；[840,1000)，则代表用户信用评分较差，风险较高。

Yule Technology collects data from anonymous cold call service scenarios with user consent, aggregates relevant information into anonymized analytical and statistical information, and processes it into a communication dimension-based credit risk scoring data product for financial service scenarios. Financial institutions, with the consent of their customers, may invoke the above-mentioned credit risk scoring data product from Yule Technology to assist in judging customers' credit risks, helping financial institutions address credit risk identification issues and improve their risk management capabilities.

The following statistical metrics are calculated: number of cold call queries in the past 35 days, proportion of hits on pan-financial labels (preferred customer groups) in the past 30 days, proportion of hits on pan-financial labels (high-risk bad customer groups) in the past 30 days, average PV (page view) of cold call queries in the past 30 days, standard deviation of cold call queries in the past 30 days, number of hits on pan-financial labels in the past 30 days, median number of hits on suspected risk labels in the past 30 days, average PV of cold call queries in the past 30 days, and standard deviation of cold call queries in the past 30 days. A scoring model is then used to generate a score measuring the probability of suspected credit risk based on these metrics.

The standard score is calculated as:
Standard score = a * number of cold call queries in the past 35 days + b * proportion of hits on pan-financial labels (preferred customer groups) in the past 30 days - c * proportion of hits on pan-financial labels (high-risk bad customer groups) in the past 30 days + (d * average PV of cold call queries in the past 30 days * standard deviation of cold call queries in the past 30 days) - (e * number of hits on pan-financial labels in the past 30 days * median number of hits on suspected risk labels in the past 30 days) + (f * average PV of cold call queries in the past 30 days * standard deviation of cold call queries in the past 30 days)

Where a, b, c, d, e, f are the weights of each indicator and can be adjusted according to actual situations.

Finally, the final score ranges correspond to the following credit risk levels:
- [0, 320): Excellent credit score, extremely low risk
- [320, 560): Good credit score, relatively low risk
- [560, 840): Average credit score, certain level of risk
- [840, 1000): Poor credit score, relatively high risk

搜集汇总

数据集介绍

特点

该数据集是基于通信维度的信用风险评分数据，规模为1亿条，每年更新，主要用于金融服务场景，通过多个统计量和算法模型评估客户信用风险。