The huge demand for credit markets and sparsely separated data have caused difficulties in the credit investigation and risk control of financial institutions. With the development of the Internet and digital technologies, qualified companies have started various credit services, and the most popular ones are the social network big data credit rating. Social credit has emerged as its own purpose, but how big is this role? It also does not seem to have reached the level of great hope that everyone has given. According to analysts in the industry, at present, social network data, as weakly variable data, has a limited role in big data credit reporting. Dr. Ding Zhuo, Chairman of Starbridge Data, a domestic startup company that uses artificial intelligence and big data to collect information, said: “Actually, social data credits account for only 5%-10% of the 360-degree portraits of users. For the assessment of the financial industry, these data cannot be used as a direct evaluation reference." Simultaneously, the intelligence report CEO Jiang Qingjun told Lei Fengwang (search for "Lei Feng Net" public concern) that the actual data associated with massive personal data and personal credit performance, the so-called Y variable, is very weak. Not easy to obtain, modeling data is not enough, of course, it is not easy to develop a mature assessment model. In addition, how high is the authenticity of data on social networks? As the main force users of social networks, we deeply understand that most of the interactions between the circle of friends, Weibo, and the state of space and comments can be attributed to the emotional “showy showâ€. Then the machine extracts credit institutions according to preset feature references. After the data you hope to obtain, the result is the real user's portrait of the object. Last August, foreign social media giant Facebook launched a patent on social big data credits. When a user applies for a loan, the lender will review the credit rating of the user's social network friend. Only if the average credit rating of these friends reaches the minimum credit requirement, the lender will continue to process the loan application. Otherwise, the application is rejected. Prior to this, Alibaba’s sesame credits launched by Ant Financial also used personal connections and consumer behavior as the basis for evaluating credit levels. In China, Zheng Haojian, general manager of Tencent Credit Co., Ltd. also elaborated on Tencent’s exploration of the construction of Internet credit information. The company mainly relied on big data and artificial intelligence technology to conduct credit information based on social data of nearly one billion users such as WeChat and QQ. , After the mining of structured data, text categorization, LBS data, social network diffusion, etc., a user portrait characterization was created. However, there seems to be a case of successful social credit reporting internationally. In the above example, Facebook's approach was criticized by the Atlantic Monthly for its "universal loan discrimination" because of its one-sidedness: Some critics believe that this patent recreates the famous "lending discrimination" practice in history: "Facebook wants to Your friend list is a reason to refuse a loan." This is also the same in the application of sesame credit's contacts, but sesame credit has collected consumer behavior data as a supplement, or it is actually the latter. As for Tencent, most of the user data legally obtained by Tencent is the behavior records of QQ and WeChat users, but these behavior records are of very low value for the use of credit information. Then, in order to achieve the purpose of evaluating individuals, Tencent may have to use the user's content data, and once it involves the user's exchange of content records, it may be considered as the legal issues of the privacy protection of the user. "Actually, for the results based on social data analysis, we can only use it as a supplement to the entire credit report, because the relationship between consumers on social networks is only loosely coupled, not like the tight Coupling formed in the corporate organization. Relationships,†Ding Zhuo explained. Everyone’s salary, flow, etc. in the company are convinced, but on social networks, content is arbitrary. "So, social network analysis can only be used as supplementary information for credit information objects other than basic data and deep data." In data types, data can be divided into strong variable data, ie credit, credit card, social security, industry and commerce, information from traditional financial institutions and government agencies, and medium-variable transaction data generated from the links of production, circulation, and consumption. And social, game and other weak variable data. "In 14,000 data sources and 700,000 data variables from different dimensions, we combined the business logic of financial institutions, analyzed the correlation coefficients of variables, and calculated the variable relationships from strong to weak decreasing, while the social network data It is obviously a weakly related variable.†Among the core team of Singbridge Data, there are 12 PhDs in computer science and finance, 27 masters, and their chief data scientist, Dr. Lin Zhenmin from Computer University of Kentucky, together with the team, the coefficients of the data variables. Years of research. Most importantly, NLP (natural language processing) is still a world-class problem. Google and Microsoft have no good solutions. Therefore, current machine learning does not have high accuracy in the processing of social data. If there is a good solution, it means that there will be some qualitative progress in the entire field of migration learning and machine learning. Similarly, current big data credits and risk control using machine learning and artificial intelligence will also improve. Dr. Ding Zhuo admits that in the 360-degree portrayal, the role of the FICO model still accounts for about 50%, and the remaining part of 45% depends on the transaction behavior data of credit investigation objects. Ding Zhuo introduced that traditional financial institutions generally rely on the FICO model for their credit ratings. However, this model is applicable to credit, credit card, foreign exchange, private lending, and other strong variable financial transaction data, “sinking to young users and young people. In the process of micro-enterprise and other customers, FICO has a lot to improve." This is also a huge price paid by Xingqiao to connect with Internet platforms such as Jingdong, Alibaba and Baidu to break the data isolation and information silos. Dr. Ding revealed that in the process of negotiations, the ability to share data resources with other large companies indirectly through third-party neutral organizations is a support for their mutual cooperation. "The significance of big data lies in its ability to calculate the correlation of massive data through machine learning, semantic analysis, and other techniques and connect them for dynamic analysis." Xingqiao’s customers include ICBC and China Merchants Bank. According to Ding Zhuo, “At present, the bank's credit customers are mainly large-scale enterprises, but in fact, they also have concerns about acquiring new customers because of sustainable development from the long-term. Look, they also need to get more customers from small and micro enterprises (supply chain finance) and young user groups (consumer finance), etc. Through cooperation, we learned that many banks also want to carry out inclusive financial services, but because of this There is less information to grasp, and they do not know how to do it. They do not understand the characteristics of the customer base of Pratt & Whitney, nor do they know how to judge the credit rating of this group of customers. Therefore, in the context of science and technology promoting financial development, FICO applicable to large enterprises also needs to have a standard to adapt to the market. “The establishment of models that are more in line with their characteristics for young users and small and micro enterprises will be something banks and credit bureaus must do in the next five years.†Optoelectronic Information Series Photoelectric information series laboratory related equipment Optoelectronic Information Product,Optical Bench Experiments Physics,Optical Devices Physics,Optical Physics Properties Yuheng Optics Co., Ltd.(Changchun) , https://www.yuhengcoder.com
