Blockchain localization cloud computing big data application evaluation method

2.1 Blockchain technology in cloud computing services

Big data is a group of data that cannot be obtained and processed in a certain period of time through traditional technical methods. It is a massive, diverse, and fast data resource that can be applied to data mining, decision-making, and other fields [8]. Cloud storage technology is a network storage service technology based on cloud computing. Through server clusters, network communication, distributed file system, and other technologies, and through the separation of software and software, it realizes the collaborative work of many various storage devices in the file server and achieves access to user data storage and services [9,10]. In short, cloud storage is a way to use cloud storage resources in the cloud for users’ storage. Users can easily access through any networkable network. The cloud computing security architecture is shown in Figure 1.

Figure 1

Cloud computing security architecture.

According to Figure 1, cloud computing security includes data security, storage and service security, and computing security. Figure 1 refers to the works of Srivastava and Khan [11]. Data security includes data confidentiality and data availability; storage and service security includes illegal user access issues, data confusion issues, data integrity issues, data duplicate storage issues, and server downtime issues; computing security includes data corruption or loss and sensitive data disclosure.

In the era of big data, terminals can no longer carry a large volume of data and store many data in cloud servers. Therefore, the security of cloud computing is an important guarantee to improve the quality of cloud computing services and attract customers, which makes big data cloud storage technology become a research hotspot in the field of big data security. In the era of big data, data storage is a promising business, but large volume of data poses a huge challenge to the traditional cloud storage system [12]. The value of big data mainly comes from its secondary use. Therefore, the integrity authentication technology of big data becomes an indispensable guarantee technology to protect the value of big data [13]. In addition, key management technology is an important technology to ensure user data security and privacy.

In terms of current distributed cloud computing, the data stored in multiple data centers are not scattered. Data are concentrated in multiple data centers. Even if one of them is invaded, it would cause a large volume of data leakage. Meanwhile, in recent years, major media around the world have continuously reported security incidents related to cloud storage, such as the disclosure of personal data of users on iCloud. However, there is no effective way to solve the security problem of distributed cloud storage.

Blockchain is a distributed database that can be added and tampered with. It is a basic technology that supports Bitcoin. In the blockchain, data ownership is established through keys, digital signatures, and account addresses, and these passwords are generated by users and stored in local files or local databases. Without another trusted administrator, the blockchain can manage its own keys. At the same time, the blockchain can also use the load verification consensus mechanism on the blockchain to attract computing resources, thus bringing greater benefits to the integrity work of nodes.

2.2 Localization of big data of blockchain technology

Localized cloud computing private network service is a new technology. Its essence is to provide users with a hybrid cloud service based on cloud computing and traditional architecture in a specific application field, including several aspects covered in Figure 2 (Figure 2 refers to the works of Manikandan and Chinnadurai [14]).

Figure 2

Application form.

2.3 IAA

Cloud computing provides an economic and effective method for the storage and analysis of massive data. Therefore, in-depth research on data mining algorithms in cloud computing has been widely used not only in theory but also in practice.

The main function of data mining is to predict the future development trend and possible behavior through the analysis of data and then make corresponding decisions. The ultimate goal of data mining is to discover hidden and meaningful information from massive data.

Data mining includes several main applications shown in Figure 3 (Figure 3 refers to the works of Yates and Islam [15]).

Figure 3

Main applications of data mining.

Concept description: A concise and typical description is obtained by describing the common characteristics and behaviors of the data set. Classification: Based on the analysis of known data, a model that can identify different types or concepts is established to predict unknown data. Clustering: Clustering analysis can be summed up as gathering birds of a feather. Its goal is to differentiate the divided groups so that each group has significant differences in characteristics in different categories, while all data in the same category have significant similarities. This process is to divide a dataset into different classes according to different attributes. Association analysis: Through mining frequent items in the dataset, the correlation between data is obtained. Prediction: By analyzing the rules and development trends of data, future data would be predicted. Isolated point analysis: This is a description of the very few, extreme and specific analysis objects and internal reasons. The discovery of outlier detection plays an important role in commercial fraud and tax evasion.

Apriori algorithm (AA) is one of the common association rule (AR) algorithms. This also includes the partition algorithm, frequent pattern growth algorithm, direct hash and pruning algorithm, and Eclat algorithm. In terms of network security, AA has been widely used in such areas as network intrusion detection. The apriori method has the following disadvantages in practical applications, especially in mining massive data.

However, the strong AR found by AA have some problems that do not conform to the reality in the process of mining, so there is some misleading in mining big data [16].

The judgment of AR in AA is only judged by the two thresholds of support and confidence. If the threshold of support is set too low, it can better mine the relevant rules, but the cost is too high, which would increase the computational complexity; if the support threshold is set too high, many useful rules would be left out, resulting in the incomplete algorithm. Both would lead to the performance degradation of AA.

AA needs to traverse the database for many times when generating frequent candidate projects. When the database size is too large, the load would increase. However, due to the slow processing speed of the system, the calculation time would increase and the calculation efficiency would decrease.

In this article, AA is optimized through the level of interest, which improves its performance. The existing interest models are analyzed and compared below.

Formula (1) shows the probability-based interest model:

In formula (1), P ( X ) is the probability of transaction X occurs in transaction set D, P ( Y ) is the probability of transaction Y occurring in D, and P ( Y ∣ X ) represents the probability of simultaneous appearance of X and Y in D. With the increase of interest Int ( X ⇒ Y ) , the mining of AR would become more and more valuable.

The variance-based interest model is shown in formula (2):

max ( Conf ( X ⇒ Y ) , Supp ( Y ) ) is the evaluation standard, and its function is to guarantee the establishment of condition ∣ Int ( X ⇒ Y ) ∣ < 1 . Conf is the dataset, and Supp is the support threshold. The combination of supportability Supp ( Y ) and confidence Conf ( X ⇒ Y ) of the latter item Y reflects the possibility that the latter item Y occurs under the influence of the antecedent item X in AR ( X ⇒ Y ) . If the difference between the confidence level of the latter item Y support Supp ( Y ) and rule ( X ⇒ Y ) is larger, and ∣ Int ( X ⇒ Y ) ∣ is larger and exceeds the set lowest interest threshold min _ int , AR ( X ⇒ Y ) has a larger value; otherwise, the value would be smaller.

The advantage of the difference-based interest model is to eliminate the impact of the high support rate of the latter item on the confidence of AR and effectively delete the rules that users are not interested in.

The correlation-based interest model is shown in formula (3):

In formula (3), Supp ( X ∪ Y ) = Count ( X ∪ Y ) / M ; Supp ( X ) = Count ( X ) / M ; Supp ( Y ) = Count ( Y ) / M . Count represents counting the number of data in a given dataset or cell range.

Int ( X ⇒ Y ) is the relationship between the former item X and the latter item Y in AR ( X ⇒ Y ) , which reflects the tightness between the former item and the latter item; among them, confidence Conf ( X ⇒ Y ) and support Supp ( X ⇒ Y ) reflect the dependence of AR ( X ⇒ Y ) and the possibility of occurrence in D.

The interest model based on correlation is defined according to the correlation between the first item X and the second item Y of AR ( X ⇒ Y ) . From the perspective of probability, the correlation between the front and rear items of AR ( X ⇒ Y ) is analyzed. If the former item X is not related to the latter item Y in probability, or the correlation is small, it means that the AR ( X ⇒ Y ) is an AR that the user is not interested in; on the contrary, AR ( X ⇒ Y ) is the AR that users are interested in.

The interest model of information content is shown in formula (4):

Among them, P ( X ) = Count ( X ) / M , P ( Y ) = Count ( Y ) / M , and P ( Y ∣ X ) = Count ( X ∪ Y ) / M .

This model combines the brevity of the correlation rule ( X ⇒ Y ) , the volume of information, and the sameness between the probability distribution of the former item X and the latter item Y, and the occurrence possibility of X P ( X ) is used to measure the conciseness of AR ( X ⇒ Y ) X. The higher the refining level of AR ( X ⇒ Y ) , the lower the frequency of X, and the higher the interest Int ( X ⇒ Y ) in the overall AR ( X ⇒ Y ) .

Formula (5) shows the influence-based benefit model:

In formula (5), Conf ( X ⇒ Y ) is the confidence level of AR ( X ⇒ Y ) ; Conf ( X ⇒ Y ¯ ) = ( M − Count ( X ∪ Y ) ) / Count ( X ) ; Supp ( Y ) = Count ( Y ) / M ; Supp ( Y ¯ ) = ( M − Count ( Y ) ) / M . Y − is the benefit calculated repeatedly between various factors.

Under the same transaction volume M and other conditions, with the increase of the number of Y occurrences Count ( Y ) , the level of interest would decline, and vice versa; as time goes on, the level of interest Int ( X ⇒ Y ) would gradually increase when Count ( X ∪ Y ) increases, and vice versa. The higher the number of occurrences of X is Count ( X ) , the lower the level of interest is Int ( X ⇒ Y ) , and vice versa.

The influence-based interest model uses the influence of X on AR ( X ⇒ Y ) to determine the level of interest of relevant rules. In this way, it can measure strong association and weak AR close to the threshold.

This article proposes a more suitable interest model by comparing and analyzing various benefit models and the actual needs of government big data mining, and the formula is as follows:

In formula (6), Conf ( X ⇒ Y ) is the confidence level of AR ( X ⇒ Y ) . In other words, this is the possibility of Y when X exists.

This article organically integrates the knowledge of probability, difference, and relevance and combines the advantages of three different interest models. In the calculation of interest level of AR ( X ⇒ Y ) , not only the coupling between the first and last two items is considered, but also the correlation between the first and last items of AR ( X ⇒ Y ) is introduced in probability. In the AR, the influence of the posterior probability on the AR is also introduced. According to the idea of differentiation, through the difference between Conf ( X ⇒ Y ) and Conf ( X ¯ ⇒ Y ) , whether the relationship between X and Y is positive or negative through the positive and negative situation of the interest Int ( X ⇒ Y ) can be known.

The level of interest calculated by Int ( X ⇒ Y ) used in the article is in the interval [1,1]. When Int ( X ⇒ Y ) > 0 , the relationship between X and Y is positive correlation; on the contrary, X is negatively correlated with Y. That is to say, X may inhibit Y. When ∣ Int ( X ⇒ Y ) ∣ > min _ int , it means that the mined AR ( X ⇒ Y ) is an AR of interest to users.

3.1 Privacy protection government data sharing framework

This article focused on the data sharing and privacy protection of government departments, including data sharing between governments. This article, from the perspective of blockchain application, fully considered the need for personal information security, and strengthened cooperation between governments, so as to promote information exchange between governments and improve the modern management level of governments. Figure 4 shows the information sharing system for government privacy protection (Figure 4 references the works of Gabelica and other scholars [17]).

Figure 4

Government data sharing framework for privacy protection.

The data sharing of the government departments is based on the internal database of the department and the joint blockchain of the department and the privacy processing module. The purpose is to build the data demand association mechanism, trust mechanism, and privacy protection mechanism between the government departments. The local database of government departments stores the data of relevant departments. Each department is responsible for the data of one or more data blocks, including data metadata, data department, time data directory, data sharing, event time, etc. In the nodes of the block, this information can clearly indicate the ownership and sharing of data sources. When problems or disputes arise in data sharing, these nodes can act as “witnesses” to deal with the rights and responsibilities of data. The “privacy processing” refers to the privacy protection measures taken to protect the user’s personal privacy before the data owner provides the data, such as anonymization, de-accurate identifier, noise interference addition, etc. The data sharing between departments adopts the “request-response” method. That is to say, through the query of the alliance blockchain, the department to which the data belongs is obtained, and the data-sharing request is sent to it. When the data provider receives the shared data, it would protect the privacy of the shared data and then submit the relevant data to the demand department, which would also store the shared data in the blockchain of the alliance.

On this basis, the government also shares information with the society. Each department publishes the information allowed by laws and regulations and protected by privacy to the government information sharing platform, and the published information includes the release time, operator, release data description, etc. The public, enterprises, and social groups can use the public information sharing platform to obtain the information needed by the government according to their own needs to achieve the purpose of social public sharing.

The government information security sharing system based on the alliance chain is proposed, and the relevant information sharing mechanism is proposed on this basis. It must be clear that when the government department obtains data, the metadata, collection time, collection department, and storage department of the data would be packaged into the data chain. It is assumed that there is no privacy leakage in the process of data acquisition and storage, and the data-sharing process between governments is further explained, as shown in Figure 5.

Figure 5

Government data sharing process of privacy protection based on blockchain.

If government department A needs information from government department B, the blockchain-based data-sharing process for public affairs between government departments can be described as follows:

The data demand department A specifies the data they want, and queries through the department alliance chain to determine the data management department B (data supply department) of the data they need.

A transmits the data-sharing request to B.

B selects appropriate data for data processing according to the data sharing request issued by A and the confidentiality policy of the affiliated unit to ensure data availability and privacy.

B transmits the confidential shared data to A and encapsulates the shared event information (such as the information of both sides of the data sharing, data description information, and event occurrence time information) into the alliance chain block.

3.2 Algorithm simulation

In order to verify the performance of the algorithm, the IAA and the CAA were simulated. Hardware in the loop simulation is the process of connecting a real controller to a fake controlled object (simulated using real-time simulation hardware) and conducting comprehensive testing of the controller in an efficient and low-cost manner. Real-time simulation hardware essentially aims to simulate the real controlled object as realistically as possible, in order to effectively deceive the controller into thinking that it is controlling a real controlled object.

The AA is a classic algorithm for mining frequent item sets that generate ARs. By using this algorithm, relationships between items can be found. The AA has two important properties: (1) a subset of frequent itemsets must be a frequent itemset. (2) The superset of a nonfrequent itemset must be a nonfrequent itemset.

To ensure the accuracy of data processing, this article uses the R data processing platform to study different big data and analyze the accuracy changes of different data calculation methods. Statistical analysis is conducted to obtain experimental results. Due to the confidentiality of government data, training data were used to simulate unexpected data: data mining. The simulation hardware environment was as follows: The CPU was Intel(R)Core(TM)i5-4250 dual-core 1.3 Hz; the memory was 6 GB; the operating system was Win764 bits. The IAA and CAA were written in JAVA language, and the development environment was jdk1.8.0_11.

Figure 6 shows the comparison of the mining time and the number of correlation rules mined by the two algorithms. In the two sets of data, the confidence threshold was set to 0.6.

Figure 6

Comparison of mining time and the number of correlation rules mined by the two algorithms: (a) mining time (s) and (b) number of mining correlation rules.

It could be seen from Figure 6a that when the support threshold of the CAA were 0.10, 0.20, 0.30, and 0.40, the mining time were 41.54, 2.84, 0.47, and 0.27 s, respectively. When the interest threshold of the IAA were set to 0.35, the mining time were 43.46, 2.75, 0.38, and 0.18 s, respectively, when the support threshold were 0.10, 0.20, 0.30, and 0.40. When the interest threshold of the IAA was set to 0.50, the mining time were 40.15, 2.82, 0.45, and 0.24 s, respectively, when the support threshold were 0.10, 0.20, 0.30, and 0.40. From the data, it could be seen that there was no great difference between the three, and the IAA not improved the mining time very well.

It could be seen from Figure 6b that when the support threshold of the CAA was 0.10, 0.20, 0.30 and 0.40, the number of mining AR was 510,548, 68,154, 16,484, and 12,048, respectively. When the interest threshold of the IAA was set to 0.35, the number of mining AR were 250,487, 62,547, 16,015, and 11,882, respectively, when the support threshold was 0.10, 0.20, 0.30, and 0.40. When the interest threshold of the IAA was set to 0.50, the number of mining AR were 175,424, 50,157, 14,315, and 10,456, respectively, when the support threshold was 0.10, 0.20, 0.30, and 0.40. It could be seen from the data that the greater the support threshold, the less AR mined by the IAA.

3.3 Performance evaluation

Figure 7 shows data related to key generation time and encryption time.

Figure 7

(a) Key generation time and (b) encryption time.

In Figure 7a, when the number of attributes of the data request user was 20, 40, 60, 80, and 100, the key generation time of the CAA was 186, 265, 335, 379, and 412 ms, respectively, and the key generation time of the IAA was 221, 298, 363, 402, and 432 ms, respectively. It could be seen from the data that with the increase in the number of data request user attributes, the key generation time would gradually approach the CAA.

In Figure 7b, when the number of encrypted files was 200, 400, 600, 800, and 1,000, the encryption time of the CAA was 312, 491, 710, 805, and 1,022 s, respectively, and the encryption time of the IAA was 76, 312, 406, 447, and 503 s, respectively. It could be seen from the data that the IAA required less encryption time than the CAA.

In conclusion, the IAA was effective and feasible in practice.

Figure 8 shows the data related to the number of query key words, trap generation time, and storage cost of the two algorithms.

Figure 8

Query key words, (a) trap generation time and (b) trap generation cost.

In Figure 8a, when the number of key words in the query was 20, 40, 60, 80, and 100, the trap generation time of the CAA was 6.22, 8.04, 8.32, 8.58, and 8.81 s, respectively, and the trap generation time of the IAA was 0.39, 0.86, 1.26, 1.76, and 2.23 s, respectively.

In Figure 8b, when the number of key words in the query was 20, 40, 60, 80, and 100, the trapdoor generation cost of the CAA was 39.87, 60.04, 72.33, 79.24, and 86.10 kB, respectively, and the trapdoor generation cost of the IAA was 4.61, 12.64, 14.02, 15.60, and 17.02 kB, respectively.

It could be seen from the data that the trapdoor generation time and trapdoor generation cost of the CAA were higher than those of the IAA. Therefore, the IAA had some advantages over GAA in trapdoor generation.

Table 1 displays the relevant data about the number of key words in the query and the search time of the results.

Table 1 shows that when the number of key words in the query were 20, 40, 60, 80, and 100, the result search time of the CAA was 1.08, 2.76, 5.15, 7.32, and 9.24 s, respectively, and the result search time of the IAA was 0.76, 1.68, 3.74, 5.69, and 7.58 s, respectively. It could be seen from the data that with the increase of the number of key words in the query, the result search time gap between the CAA and the IAA was getting larger and larger. The result search time of the IAA was shorter than that of the CAA.

Table 2 shows the relevant data about the number of key words in the query and the search time of the results.

Table 2 suggests that when the number of key words in the query was 20, 40, 60, 80, and 100, the result search cost of the CAA was 12.43, 26.84, 50.54, 71.34, and 91.55 kB, respectively, and the result search time of the IAA was 5.05, 15.21, 21.45, 37.62, and 63.72 kB, respectively. It could be seen from the data that with the increase of the number of key words in the query, the gap between the result search cost of the CAA and the IAA was becoming larger and larger. The result search cost of the IAA was lower than that of the CAA.