Intelligent evaluation method of bank digital transformation credibility based on big data analysis

2.1 Digital transformation strategy of banks

The digital transformation strategy uses the thinking of bank reengineering theory to carry out around the bank culture, embracing financial technology, focusing on customer experience and reiterating the bank’s operation and management system:

(1)

From customer to user, restate service object and business thinking: The bank has broken through the existing service boundary, expanded from the customer system with bank card account as the core to class II and III accounts, and app users without bank card account, striving to build an Internet funnel-shaped user system. The bank has changed its thinking from transaction to user experience. Guided by this, it adheres to the business philosophy of strengthening the Polaris index, that is, monthly active users [9]. From the customer’s point of view, it creates the best customer experience bank, innovates the bank’s products and services, and is committed to interaction design. Taking the user experience as the strategic commanding point, we first established a system to monitor the user experience, then recorded the customer’s feelings in real time, and finally quickly fed back and improved. We always provided support for the online service platform and customer manager through intelligent way, and fundamentally improved the customer experience.

(2)

From bank card to app, reiterate the bank service mode: With the change of customer behavior, the main position of bank and customer transaction is app. Bank card is only a product to serve customers, but app is a service platform [10]. Taking “China Merchants Bank” and “palm life” as platforms, self built consumption scenarios and external expansion scenarios, used API interface to transmit financial services, strengthened the universality of ecological alliance construction, and initially built a convenient ecosystem for travel scenarios, including public transportation, subway, parking lot, etc.

(3)

From centralization to opening up, the bank reiterated its scientific and technological strength and corporate culture: The most basic support of banks is science and technology, through financial technology investment benchmarking financial technology companies, build an open its system, and comprehensively strengthen the R&D and application of science and technology. The amount of the bank’s financial technology innovation project fund increased from 1% of the pre-tax profit in 2017 to 1% of the operating income of the previous year, and established a financial technology innovation incubation platform, established an independent operation mechanism, and increased the reserve of professional technology and data personnel [11]. In addition, the company’s articles of association were revised and it was proposed that in principle, the total budget amount of financial technology investment each year should not be less than 3.5% of the operating revenue after the audit of the previous year. Culture is at the bottom of financial technology. It launched “eggshell”, an exchange platform for employees, and launched an “open, flat and inclusive” Internet corporate culture. Through the establishment of fault tolerance mechanism, it supports all kinds of innovation and is committed to changing the traditional banking bureaucratic culture.

(4)

From organization to action, the bank organization construction and management upgrading are reiterated: Banks began to form more task-based teams, from organization to action [12]. First of all, gradually build an enabling organization, and strive to improve the service ability of the middle station. Secondly, we should use financial technology to get through all business systems. With massive data as the core, we should establish the data center, integrate internal and external data, improve the big data management system, and finally build the “general staff”. Strengthen the construction of business platform, and reduce the burden on the front line by empowering the market. In general, the middle platform enables the front desk and the front desk feedback to accelerate the upgrading of the middle desk, so as to promote the self evolution and upgrading of the organization and management.

2.2 Introduction of big data analysis technology

The existing credibility evaluation methods have the shortcomings of backward data processing technology. In order to make up for the shortcomings of the existing methods, this study introduces big data analysis technology to improve the validity of the credibility intelligent evaluation results of bank digital transformation.

There are many kinds of big data analysis technology. According to the method requirements, the introduction of online analytical processing technology is a typical data cube technology. It extracts effective information from the data cube through appropriate transformation, so as to carry out subsequent data mining and analysis work [13]. Common OLAP operations have the following forms: roll up, that is, to coarsen the data granularity in the same dimension appropriately, and gather the data to a higher concept layer; drill down, that is, to refine or supplement the data granularity in the same dimension, and provide some information, so as to disperse the data to a lower concept layer; cut block, that is, to select a part of the data cube and separate it. In order to observe and analyze the other two dimensions, we need to list the sub cubes separately; in slice, we need to set one dimension of the data cube as a fixed value, and simplify the data cube into a data square; in axis, we need to rotate the data cube or its blocks or slices by an angle, or change the relative position of some dimensions, so as to provide different perspectives for observation and analysis.

Next, take a bank as an example to analyze the data cube and OLAP technology. The three dimensions of the data cube based on the bank’s profit are time, address and service, and the value of the cube is profit. The OLAP operation of the data cube is as Fig. 1.

OPEN IN VIEWER

In addition, because the digital transformation of banks will produce a large amount of data, a series of simple data preprocessing technologies are needed, such as data cleaning, data integration, data specification, data transformation, etc. [14].

Data cleaning, through the preliminary browsing and analysis of the original data, the error points, omissions and noise in the data are screened out, and the impact of the poor quality of the data on data mining is minimized. There are three common methods for data cleaning: delete, modify and fit. For the data with serious errors and omissions, which contain almost no effective information, and the obvious noise and outliers, it is recommended to delete directly; however, if the data is indispensable in database construction and subsequent analysis, it is recommended to contact the first-line department to provide data again or recollect the data by means of field means. For the data with obvious errors, but still can distinguish or extract the correct information from the data, it is suggested to revise the data, such as “20020” to “2020”. If there are some errors or omissions in the data, but the trend of data change can be estimated from the data before and after, the data can be fitted by certain mathematical means. The common data fitting methods include: using center measurement, global constant to fill in the defect value, or fitting the defect value by linear function, multivariate function, Bayesian network and decision tree.

Data integration, entity identification of multi-source heterogeneous data through relevant means, and the identified data are classified and integrated. Entity recognition is a complex and systematic project; in the digital transformation data of banks, how to extract consistent information from words or data with diverse expressions and complex and confusing formats, the common processing methods are as follows: Firstly, keyboard input instead of manual handwriting, which can largely avoid information reading errors caused by different fonts or scribbled handwriting; Secondly, keyboard input instead of manual handwriting. The information is filtered by keywords; Thirdly, the relevant information is coded uniformly. If information is to be input in the system, the system will remind the user to select the appropriate code in the drop-down menu, which can largely avoid the confusion caused by different descriptions of the same object or event [15].

Data specification reduces the number of random variables or attributes by mathematical means to simplify the data structure and retain the intrinsic information to the greatest extent [16]. The common ways of data specification are attribute deletion, attribute generalization and attribute construction. Data transformation, by merging and adjusting the quantity or format of the original data, makes the presentation of data more in line with the psychological expectations of decision makers, and provides a better perspective for data observation and analysis [17]. Common data transformation methods include normalization and discretization. Normalization, by merging the data into a unified interval, such as [ 0 , 1 ] or [ - 1 , 1 ] , to avoid the impact of dimension and value range differences on data analysis. At present, linear method is commonly used to realize data normalization:

x i ⁢ j = x i ⁢ j ′ - min ⁡ x j ′ max ⁡ x j ′ - min ⁡ x j ′(1)

In Eq. (1), x i ⁢ j ′ represents the original value of the j index of the i object; max ⁡ x j ′ represents the maximum original value of the j index of all objects; min ⁡ x j ′ represents the minimum original value of the j index of all objects; x i ⁢ j represents the normalized value of the j index of the i object.

Based on the above analysis results of bank digital transformation strategy and the introduction of big data analysis technology, select credibility intelligent evaluation index, build credibility evaluation model, and lay a solid foundation for the realization of credibility intelligent evaluation of bank digital transformation [18].

According to the national bank data standards, credibility refers to the ability of the system to perform functions when needed, including reliability, availability, recovery, maintainability and other aspects. The credibility of an event is defined as the average of probability and necessity, and the expression is:

C ⁢ r ⁢ { A } = 1 2 ⁢ ( 𝑃𝑜𝑠 ⁢ { A } + 𝑁𝑒𝑐 ⁢ { A } )(2)

In Eq. (2), C ⁢ r ⁢ { A } represents the credibility measure of event A , 𝑃𝑜𝑠 ⁢ { A } and 𝑁𝑒𝑐 ⁢ { A } represents the possibility space and necessity space respectively.

In addition, the credibility measurement between data sources is also a key part of the credibility intelligent evaluation of bank digital transformation. Its main content is to build the calculation model of local credibility and global credibility between data sources [19]. The credibility between data sources refers to the comprehensive credibility of the local data sources to the target data sources, which is restricted by the local credibility and the global credibility between data sources.

This chapter studies the status of local credibility and global credibility in the model, the selection of parameter values, the influence of data interaction, the measurement of similarity recognition between two data sources, and the influence of time on the model, and puts forward a new method of credibility between data sources.

In the process of credibility intelligent evaluation of bank digital transformation, when the credibility value between two data sources exceeds a certain threshold, a directed link is established between the two data sources. With the expansion of network scale, trusted virtual network is becoming more and more stable. If the data provided by a data source is found to be untrustworthy or malicious, the model can quickly impose a penalty coefficient on the provider, which will reduce the credibility of the provider in a period of time. But as time goes on, if the data source can continue to provide reliable data, its credibility will be slowly restored [20]. In credibility analysis network, if there is no new context interaction between data sources in a computing interval, a time penalty is imposed on them.

Through the above process, the construction of credibility evaluation model is completed, which provides the basis for the following credibility evaluation steps, and also prepares for the final credibility evaluation [21, 22].

According to the above-mentioned credibility intelligent evaluation model, the specific implementation steps of the credibility intelligent evaluation are formulated as follows:

Step 1: Data aggregation and information extraction.

After the original data of digital transformation of banks is preprocessed in the data dictionary format described in Section 2.2, the data will be stored in the form of items with a fine granularity. In order to observe and evaluate the digital state of banks, we must aggregate and improve the data items in order to extract the feature information from them. The common data aggregation methods include the on-line analysis and data protocol technology in data preprocessing. In volume, the data of the same dimension will rise along the conceptual level, so that the data can be aggregated to a higher level. Data specification, that is, through the technology of attribute deletion, attribute generalization and attribute construction, it simplifies the data structure and retains the feature information. The two technologies are in a sense identical and can be used in combination [23].

For the aggregation of bank digital transformation data, the first thing to consider is the volume up in the space-time dimension. After repeated debugging and demonstration, the location information should be rolled up to the line level, and the time information should be rolled up to the quarter level. In other words, the credibility evaluation of OCS should take the bank digital transformation of each line every quarter as the evaluation object [24].

For the quality evaluation index, it is suggested to extract it through the dictionary of OCS fault parameters. The extraction method is as follows: set the fault parameter importance code to “2”, then the bank digital transformation important parameters can be screened out. It should be noted that there are 10 important parameters specified in the fault parameter dictionary, while only 7 secondary indexes are specified in the quality evaluation, so the above 10 parameters need to be integrated to make them correspond to the 7 indexes of the quality evaluation.

After the preliminary extraction of quality evaluation indicators, the indicators should be rolled up to the line level along the two dimensions of time and space. As for how to realize the aggregation of parameter values in the process of roll up, it is suggested to keep the parameter extremum without considering the average value [25, 26]. This is because the quality evaluation mainly focuses on the health status of the bank’s digital transformation under extreme parameters. If the bank’s digital transformation can withstand the influence of extreme parameters, there will be no problems under moderate parameters; if the evaluation is based on the mean value of parameters, even if the evaluation results are good, it can not guarantee the normal operation of the bank’s digital transformation under extreme parameters.

For the quality appraisal index, availability index and maintainability index, because they cannot be directly obtained from the data entry, it is necessary to first roll up the data to the line level, and then construct the attribute according to the definition. For example, after the fault items are rolled up, the fault frequency of a certain line in a certain quarter can be counted, and then divided by the total length of the line to get the fault intensity [27]. Other attributes are constructed in this way and so on.

Step 2: Normalization of index parameters.

The linear normalization method is recommended, but it is only suitable for the situation that the larger the index value, the better. According to the essential characteristics and evaluation requirements of bank digital transformation data indicators, this paper makes appropriate improvements to the method. The improved method divides the indicators in the digital transformation data of banks into the following major types: very large indicators, that is, the larger the index value is, the better; very small indicators, that is, the smaller the index value is, the better; interval indicators, that is, the index value should fall into a certain interval [28]. The interval classification of each index should be carried out in strict accordance with the standard.

Step 3: Determining the index weight.

Considering that the current digital transformation data management system of China’s banks needs to be optimized and improved, there are a large number of errors and omissions in the data collected by the existing technology, we still need to learn from the existing experience of bank digital transformation operation and maintenance to correct and supplement the errors and deficiencies of objective data. Therefore, in the use of AHP, the experience driven AHP should be the main method, and the data driven AHP should be the auxiliary method.

Step 4: Score calculation and grading.

It is recommended to use the weight obtained by AHP moisture weight method to directly sum the normalized values of each index to obtain the overall credibility score of the digital transformation object of the bank.

Step 5: Validity test of evaluation results.

It is recommended to combine Kendell harmony coefficient with Spearman rank correlation coefficient.

Through the above process, the intelligent evaluation of the credibility of the digital transformation of banks is realized, which provides accurate data support and help for the digital transformation of banks, and also provides some reference for the research of the credibility intelligent evaluation.

3.1 Selection of subjects

In order to simplify the experimental process, we select a local bank as the object to collect its digital transformation data.

3.2 Determination of evaluation validity test method

The validity test of evaluation results mainly uses two methods: Kendell harmony coefficient method and Spearman rank correlation coefficient method. Among them, Kendell harmony coefficient method can calculate the similarity of the series generated by two or more evaluation methods, which is suitable for the preliminary test of the evaluation results; Spearman rank correlation coefficient method can be used to generate the benchmark grade series, and then the rank correlation coefficient can be obtained from the evaluation results of several methods and the benchmark grade series. The higher the rank correlation coefficient is, the higher the validity of the evaluation method is, which is suitable for the preliminary test of the evaluation results. The results were further tested. According to the experimental requirements, Kendell harmony coefficient method is selected.

Kendell harmony coefficient method is to add different ranking values of the same object in three ranking sequences generated by AHP method, grey clustering method and TOPSIS method to obtain the sum of ranking values R i and the square R i 2 of the sum of ranking values. The Kendell harmony coefficient can be calculated by substituting the calculation results into the following formula:

W = 12 ⁢ [ n ⁢ ∑ i = 1 n R i 2 - ( ∑ i = 1 n R i ) 2 ] k 2 ⁢ n 2 ⁢ ( n + 1 ) ⁢ ( n - 1 )(3)

In Eq. (3), k is the number of evaluation methods; n is the number of evaluation objects.

According to the selected experimental objects, the evaluation validity test method is determined to evaluate the credibility of the digital transformation of banks. The validity of the evaluation results shows the application performance of the method. Under normal circumstances, the higher the validity of the evaluation results, the better the comprehensiveness of the evaluation, that is, the more accurate the evaluation results. The validity data of the evaluation results are as Table 1.

As shown in Table 1, compared with the existing methods, the validity of the evaluation results of the proposed method is greater, indicating that the proposed method is more comprehensive and accurate, which fully confirms the effectiveness and feasibility of the proposed method.