Big data audit based on financial sharing service model

2.1.1 The task of data collection is heavy

In the traditional financial accounting mode, the financial information inside the unit is not highly related, and it is difficult to communicate with each other effectively. If it is necessary to audit the unit, it is necessary to integrate the accounting information of each department into the relevant audit data of the unit, which greatly increases the work task. However, the above difficulties will not be encountered in the financial sharing service mode. With the cloud platform system, auditors can easily obtain the relevant data required by the audit, saving a lot of human and material costs.

2.1.2 Heavy task of data sorting

Under the traditional financial accounting model, there is no effective connection between relevant departments of the unit, and there are many differences in data interfaces and standards. This requires auditors to carefully collect relevant information on the one hand and to sort out the data on the other hand. The length of the sorting process and the complexity of tasks greatly reduce audit efficiency and the audit cost remains high. In the financial shared service model, the establishment of the cloud platform system has standardized various types of financial data, and the related business systems have also been organically connected. Audit staff do not need to consider how to spend too much manpower and material to organize data when collecting standardized data, and the quality of the data can be better guaranteed [5].

2.1.3 Inefficient use of external data

In the traditional financial accounting mode, because the accounting information system of each department and the relevant business system are relatively independent from each other, it is difficult for auditors to collect all kinds of external information related to audit in each business system. However, under the financial sharing service mode, the cloud platform system can realize the effective management of internal and external information of the enterprise. Through this platform system, the audit staff can easily obtain the required external data, which will greatly improve the efficiency of data collection.

2.2.1 Big data collection

The various types of data collection based on the cloud platform system are mainly conducted in the following three ways: The first method is to directly collect financial data, business data and various strategic data related to audit objectives through the cloud platform system; the second way is to complete the seamless connection with the cloud platform system. For example, from the ERP system, the auditors can directly obtain all kinds of relevant data directly related to the production and sales of the enterprise. The third way is to obtain various types of data related to enterprise audits from outside the enterprise [6]. Under the financial shared service model, standardized financial data can reduce the work pressure of auditors during the data collection process, and can also reduce audit costs and improve audit efficiency.

2.2.2 Big data cleaning

When the big data collection is completed, the relevant data needs to be cleaned in order to further improve the quality of the audit related data, and at the same time correct or eliminate any conflicts in the various types of data. Although under the financial shared service model, various types of financial data or other data in the cloud platform system have been standardized, but in order to improve the quality of the data, the relevant data still needs to be cleaned, which is a process that preprocessing must complete [7].

2.2.3 Store big data

After completing the cleaning of big data, the auditors need to establish an audit database to centrally manage and store the various types of cleaned data so that the audit staff can extract the relevant data at any time.

3.2.1 Determine the audit objectives

Essentially, big data auditing is an auditing method. This auditing method has no effect on the audit staff in setting audit goals. The determination of the audit objective requires the audit staff to choose according to the actual situation. From the perspective of the executive, we can see that the big data audit based on the financial shared service model can not only serve the internal audit of the internal auditors of the enterprise, but also provide the external audit services of the auditors of external accounting firms to the enterprise [9]. From the perspective of audit objects, on the one hand, big data auditing can audit the business of the enterprise, and on the other hand, it can audit the accounting information of the cloud platform system. If the object of the audit is all kinds of business of the enterprise, the focus of the audit should be on the process and capital flow that can fully reflect the business. If the object of the audit is a cloud platform information system, the focus of the audit should be on whether the cloud platform system can protect enterprise assets, maintain data, and minimize the consumption of economic sources. If an internal audit is required, the focus of the audit should be on whether the financial information system based on the shared financial service model can help management decisions. If an external audit is required, the focus of the audit should be on judging whether the various financial statements provided by the enterprise to the expected information users in the financial shared service model can help its decision-making [10].

3.2.2 Identify and evaluate audit risks

The importance has a certain negative impact on the risk of big data audits. The process and results of big data pre-processing will also have a certain negative impact on the audit in various ways and bring certain audit risks. Normally, if the pretreatment process fails to comply with the regulations, or the audit results and audit requirements are different, it will increase the probability of audit risk [13]. In addition, audit risk is also closely related to the credibility of the cloud platform system. If the credibility is low, on the one hand, it will increase the audit risk, and on the other hand, it will also have a certain impact on the audit results. In order to effectively identify and evaluate audit risks, it is necessary to effectively supervise the big data preprocessing process. The audit working group should also use a professional third party to evaluate the internal control and credibility of the cloud platform system and other systems, and use this to judge the status of the audit risk.

3.2.3 Develop and improve the audit plan

When formulating a big data audit plan, the scope, time, arrangement of specific matters, and resources required for the audit should be fully taken into account. In particular, attention should be paid to the time and labor arrangements related to the big data preprocessing process. Compared with traditional audit methods, the audit scope specified by big data audit is wider. These data mainly come from the audit database constructed by the previous big data preprocessing link, which contains internal and external financial data related to the audit and other data related to the audit.

3.2.4 Selection of operational audit procedures

The big data audit procedure based on the financial shared service model requires the following three steps to achieve: Data Analysis, Establishment of Intermediate Tables and Formation of Audit doubts. About the technical principles of these tools, the article will be specifically discussed in Chapter 5.

3.2.5 Implementation of audit procedures

According to the audit plan, we should start to implement the relevant audit work. In the specific implementation process, we should conduct a comprehensive supervision on the collection, sorting and storage of all kinds of audit big data. Third party experts have conducted professional comprehensive evaluation on the reliability of “cloud platform system” and other business systems. Audit staff should take this as the basis, find evidence through cloud platform system and other business systems to confirm audit doubts, and lay a good foundation for the formation of audit results. The problems found in the audit process should be fed back to the management in time, and the review and evaluation of these problems should be done well.

3.2.6 Report audit results to management in a timely manner

The auditors determine the doubtful points of the audit according to the audit plan, and obtain various audit evidences based on them to confirm the doubtful points of the audit and obtain the audit results. Before making the final audit conclusion, the auditor needs to fully consider the auditee and its environment, and should also refer to the previously determined audit objectives. All kinds of misreporting problems found in the audit process should be reported to the audited management in a timely manner, and specific improvement opinions can also be given, and an audit report will be formed after the management’s response.

4.1.1 Big data mining technology

Data analysis and mining is the most critical step in big data processing. Data analysis mainly uses big data analysis tools to analyze and classify the massive data stored in the distributed database or distributed computing cluster to meet the common analysis needs. For example, some real-time requirements will use EMC’s Greenplum, Oracle’s exadata, and MySQL based columnar storage infobright. For some requirements based on semi-structured data or batch processing, Hadoop can be used in the process of statistical analysis. Because of the large amount of data involved, it will take up a lot of system resources, especially I / O. In addition, there are high requirements for the use of statistical tools and keywords that need to be classified. They determine whether the data can be classified accurately, which will directly affect the accuracy of data mining value. Data mining is based on the calculation of various data mining algorithms for the big data that has done a good job in statistics. It extracts the hidden and potentially significant information, reveals its laws and results, and is used for decision-making and prediction. Therefore, the traditional relational, structured data sets and mining methods are no longer applicable. Big data mining generally does not have a pre-set topic, only accurate and appropriate algorithm can get valuable data analysis results. The typical algorithms are K-means for clustering, SVM for statistical learning and naive Bayes for classification. The main tools used are mahout of Hadoop.

Auditors can use automatic big data mining algorithm to deal with all kinds of big data in data warehouse, so as to find out all kinds of problems hidden in all kinds of data, and finally determine audit doubts. The main functions of big data audit data mining technology based on financial sharing service mode are as follows: on the one hand, it can discover the financial, non-financial or other business data abnormalities through this technology; on the other hand, auditors can associate the financial data in cloud platform system with business system data such as ERP. So as to establish the cross checking relationship between various types of data, and then determine the audit doubts, and find out the problems in the business work [14].

The application of big data mining technology mainly adopts Apriori algorithm, which is introduced as follows.

Apriori algorithm is a frequent item set algorithm for mining association rules. Apriori algorithm scans transaction database many times, and generates frequent sets by using candidate frequent sets every time. Its main steps can be divided into defining minimum support, filtering all frequent item sets and generating association rules based on confidence. Its algorithm is mainly based on the distributed implementation of spark, as shown in Fig. 2 [15].

OPEN IN VIEWER

Fig. 2

Implementation flow chart of distributed Apriori algorithm.

The main idea of the algorithm is as follows:

Step 1: Generate frequent 1 item set L₁. The transaction set T is distributed to each machine in the form of RDD < string, l > . The cumulative number of items and the item set with higher support are reserved.

From frequent k term set L_k, frequent K + 1 term set L_k +1 is generated. LK self-connects to generate C_K +1, and then scans the database to generate L_K +1 according to C + 1.

The core steps of parallel Apriori algorithm are:

val fiml = transactions. FlatMap(1ine = >line). Map ((,1))

ReduceByKey (_ + _). Filter (_._ 2 > min-Support)

savefiml (fiml,output+”result--1”,sum)

def savefiml(fim:RDD[(string,int)],path:string,

count:double):Unit = {

fim.map(1ine = >{

line._1+“:%.2f”.format(1ine.–2/count)}).

Coalesce(1, true). SaveAsTextFile(path)}

var(trans,newfim)  = LI tofim2(rim I.rnap(_-____________1).

    collect,transactions,minSupport)

save(newfim,output+”resultÒ2”,sum)

def

L 1 tofim2(L 1:Array[String],trans:RDD[(String,

Int)],minSupport:Double):RDD[(List[String],Int)]  = {

val L 1 C—L 1.size

val citems = scala.collection.mutable.ArrayBuffer[List

[String]]()

for(i< -0 until Llc){

for(j< -i+1 until Llc){

Citems + = List(L 1(i), L1(j)). sortWith(_ < _)

}

}

Step 2: Supporting data De duplication data holding certification scheme:

If the user is logging in to the data client for the first time, the first step is to register an account, which contains a unique user name, ID and password. At the same time, all the information in the account will also be owned by the cloud service provider and the trusted third party.

At this point, the key used to validate the file is generated by calculation on the local client.

The specific steps are: to find the large prime number p,q, when p = 2p + 1 and q = 2q + 1, and meet the condition of N = pq at the same time, the generator g needs to be found in the category of the positive integer PQ cyclic subgroup, and then find the large prime number e and large prime number d that are larger than the safety parameters, so that the large prime number e and large prime number d meet ed = 1(mod pq), and then the binary number v of k-bit will be randomly generated by the system.

Among them, the private key sk is embodied by (e,d,v) and the public key pk is embodied by (N,g). The above steps are the same as the traditional PDP scheme. After the public key and private key are generated, the new password PW needs to be input. After the private key is encrypted by AES, it is stored again, and then the public key is publicly stored in the database. The user needs to upload the file F. the client first calculates the MD5 value M of the file, and then sends the result to the cloud server after the calculation. The verification process of the cloud server for the received M is as follows:

if (searchHash (HashTable,m)==true):

sendClien (t randomKey,FID);

else requestFileTag ()

In the process of searching hash table by the server, if M can be verified, the system will generate a random ran domkey key after returning to the client, and the system will continue to run step (2) after the key generation is completed. However, if the existence of M cannot be verified, the system will request the client to upload the file label and the corresponding file. (2) When the server finds the existence of m in the running process, the system returns to the client and generates a file number FID and a random key.

The client sends the result to the cloud server through the calculation of M ‘value and FID of the file. The secondary verification process of the received M’ by the cloud server is as follows:

F = getFile (FID);

temp = MAC (F, randomKey);

if (m^′==temp):

addFileUser (FID,user);

sendThirdParty (FID);

else requestFileTag ();

The client of the system includes user and file management, visual verification and integrity verification. The user management business in the client is realized through the user management module, which mainly includes two businesses: user login logic and user registration. The file management module implements all business logic related to file storage in cloud storage. The visualization of the client interface is realized by the visualization module. At the same time, the module supports the user to read, write and manage the file on the client, which can give the user a better practical embodiment. Using the client to verify the integrity of files can improve the business logic, and then effectively complete the verification module.

The MD / MF (multi-dimensional / multi feature analysis) model, which was established by corpus linguist Douglas Biber in 1988, is used to describe the discourse relationship between spoken and written styles. This method uses standardized computer-based text corpus and automatic recognition technology to calculate frequent lexical and grammatical features. The co-occurrence pattern of these features can be used to analyze and confirm the functional dimensions of language variation in texts through multivariate statistical methods, and provide a comprehensive description of the relationship between the genres of these dimensions.

This is a common analysis method in big data audit. Based on the multidimensional data stored in the audit data warehouse, the audit staff summarize, associate, cluster, classify and analyze it from different perspectives. Compared with data mining technology, multi-dimensional analysis method is more convenient to find problems in business process in the process of data analysis, and obtain audit doubts. Finally, the audit staff get the audit evidence and form the objective audit conclusion with the help of the multi-dimensional data aggregate audit. The algorithm formula is as follows: C f ( α , β ) = Cor ( β • f ( x ) ⊕ α • x ) … …(1)

If C _f  (α, β)  = 0, the linear approximation is called a zero-correlation linear approximation.

For each cipher text selected, it is obtained by partial encryption and decryption (p, c), the value of z can be obtained by formula (1), and the counter N [z] is established.

T = ∑ Z = 0 2 m - 1 ( N [ z ] - N 2 - m ) 2 N 2 - m ( 1 - 2 - m ) ≈ N 2 m ∑ Z = 0 2 m - 1 ( N [ z ] N - 1 2 m ) … …(2)

If the guessed key is the correct key, then the statistic T is expected to be μ 0 = ( ℓ - 1 ) 2 n - N 2 n - 1 … …(3)

Then the formula can be deduced N = ( 2 n - 1 ) ( z 1 - η 1 + z 1 - η 2 ) ( ℓ - 1 ) / 2 + z 1 - η 1 … …(4)

In addition to the above two methods, SQL query technology is also a frequently used data analysis method in big data audits. SQL query technology can achieve cross query and fuzzy query between multiple tables. Audit staff can use this technology to find all kinds of potential problems in business processes faster and determine audit doubts. In order to verify the technical effects of SQL, this article compares it with traditional methods. The empirical results show that SQL search has a significant cost difference compared to traditional methods. See Figs. 3 and 4.

OPEN IN VIEWER

Fig. 3

Traditional way of query.

OPEN IN VIEWER

Fig. 4

SQL query.