A Crypto Scheme Using Data Obfuscation of Entity Detection and Replacement for Private Cloud

Advancement of Technology from over the years has come a long way. One of such was the idea of cloud computing. It helps to solve the problems of traditional data storage systems. Its cost is very less when compared to other sources and easy to use. The client pays as he uses cloud assets [1]. Majorly it is profitable for those who have tiny enterprises and self-experts who expand their applications without having large investments. It economically eases as he pays what he uses. There are three modules where this technology mainly works. The primary is service providing person [2], who is responsible for letting information be kept in the cloud. A secondary module is a consumer who can have permission to access the essential services from the cloud. And the last is an interface of the web by which users can keep resources in the cloud and even can retrieve the services that are required [3]. These services are mainly classified as software as a service (SaaS), infrastructure as a service (IaaS), Application as a service (AaaS), and Platform as a service (PaaS), Communication as a service, Network as a service and Database as a service [4]. In the cloud, these services will be deployed through private, public, and hybrid models [5]. Cloud has very significant features that differentiate it from other technologies. They are pay-as-you-go [5], auto-scaling [6], flexibility [7], elasticity. According to the pay-as-you-go process of billing is done based on the utilization of the amount of the resources. This is playing an essential role in the upward of this technology.

Regardless of having such more prominent features, it thinks of security issues [7]. The security issues are associated with data [8], network, infrastructure, privacy, assaults, and so on [9, 10]. The significant research work concentrated on giving security to information in the cloud from being accessed by unauthorized clients [11]. This is known as data/information security [12]. Data that is put away in the cloud ought to be kept private as per the necessities of the cloud client. Cloud still not ensured data security [13, 14]. Information that is placed/shared through cloud commonly incorporates client transaction data, client inclinations, and patient's health records, etc. Subsequently, applications in cloud research ordinarily require compelling strategies/mechanisms to protect the security of the cloud client's sensitive data [15]. These strategies/mechanisms should work viably by appropriately handling, testing, and examining the huge amount of data. Real-time information, which has the most sensitivity, is generally attractive for checking and approving the viability and appropriateness of these strategies and mechanisms.

There is a need for strategies and mechanisms that are viable and effective to guarantee the protection of such client data before it is made open [16, 17]. A few methods that scramble or supplant the delicate information in reports in a way that doesn't uncover the sensitive or the private data. These methods guarantee that when files are shared, the ill-conceived client couldn't catch the client's delicate data yet preserves the readability of information [18]. Other methods incorporate cryptography, which converts the information into some other form that can't be read and understand by anybody except if it is transformed into its original form [19, 20]. In the principal case, however, it doesn't give genuine data, there is as yet a chance of information breach since it permits the readability. In the subsequent case, however, it doesn't empower the readability, there is a likelihood that outsider can decrypt the file and use it. Thus, mechanisms ought to be planned with the end goal that no secret data is uncovered to the ill-conceived client, yet the file ought to be in a way that other analytical process can be performed on the file [14, 21, 22].

In our work, we designed and implemented an effective crypto-scheme for preserving the privacy of unstructured data such as company policies [23]. Our mechanism protects the data in two phases. In the first phase, it replaces the first data with sham data by detecting and supplanting the keywords in the file using Natural Language Processing (NLP) methods [24, 25] and permits the readability of the file to such an extent that it is conceivable to reproduce the original file. Retaining the readability of the file has two advantages. (I) when an outsider (ill-conceived client) get to gain access to file, he can't conclude which information is original and which is supplanted, (ii) since enough data accessible, we can perform analytical and processing Application on file. In the second phase, it transforms the delicate data into an unreadable form using a conventional RSA algorithm.

The overview of the paper is as follows. Section-II shows the related works done on ensuring data security in cloud computing. The working of proposed scheme has been described in section-III. Section-IV provides experimental results of proposed scheme with sample data. Section-V ends with the conclusion.

2.1.png

2.2.png

Figure 2. Workflow of mechanism

The proposed algorithm has three modules, Obfuscator and Cryptanalysis module.

Obfuscator: This module plays out the obscurity and de-obscurity operations on an input file that contains unstructured information. At whatever point a client needs to store his delicate data in the cloud, the obfuscator module scramble the data in the file and sends it to the crypt-analysis module. Furthermore, de-scramble the file, at whatever point client needs to use it. This module endeavors to holds the readability of the file to such an extent that it is conceivable to reproduce the original file [28, 29].

Cryptanalysis: Usually cryptanalysis module transforms the information into a form that can't be read and understand by other users who are not authorized. It takes the obfuscated document and encrypts it by using a public key encryption algorithm RSA, which can be implemented practically and yields better results than the rest of the algorithms [30, 31]. The workflow algorithms are shown in Figure 2.

Initially, the sample data is collected for obfuscation. To extract the keywords from the document we perform the following steps.

Data pre-processing: Data pre-processing includes two steps they are (i) Noise removal and (ii) Normalization.

Noise removal: Various data pre-processing techniques have been applied to the data for noise removal. Techniques have been used to remove the redundant keywords, punctuations, URLs, tags, and stop words, etc. [25]. Stop words refer to a large number of prepositions, which are to be filtered out before processing the data.

Normalization: Normalization is a process of converting a list of words to a more uniform sequence. This is useful in preparing text for later processing. Techniques like converting the text into lower case, parts of speech (POS) tagging, lemmatization has been applied on the data for smooth processing. Lemmatization and POS tagging are the special cases of normalization [30].

Lemmatization: Lemmatization is a more advanced technique that works based on the root of the word.

POS tagging: It is the way toward labeling a word in a corpus to a corresponding part of a speech tag, in accordance with its context and definition. POS Tagging is a significant step to comprehend the meaning of any sentence or to extricate a relationship and fabricate a knowledge graph. Python NLP bundle is utilized for actualizing the procedures.

Lexical analysis: It includes recognizing and investigating the structure of words. The vocabulary of a language implies the assortment of words and expressions in a language. The lexical examination is separating the entire lump of content into passages, sentences, and words [31].

Topic based modelling: Topic vocabulary has been built using Latent Dirichlet Allocation (LDA) model. Based on the number of topics constructed from LDA, a set of seed words will be extracted for building the Topic-Based Modeling.

A set of words for each topic can be extracted by using the below parameters.

$P\left(\theta_{1: M} Z_{1: M} \beta_{1: k} / D ; \alpha_{1: M}, \eta_{1: k}\right.$

where, $' \theta^{\prime}$ denote the distribution of topics, ${ }^{\prime} z^{\prime}$ a number of topics for each report ${ }^{\prime} \beta^{\prime}$ signify the dispersion of words for every topic. Given ${ }^{\prime} D^{\prime}$ denotes the document ${ }^{\prime} \alpha^{\prime}$ and $' \eta^{\prime}$ denotes the parameter vectors for document and topic, respectively.

But this can't be calculated nicely because this entity is intractable. To solve this problem, variational inference technique has been used to minimize the Kullback-Leibler divergence (KL divergence) between the approximation and true posterior as an optimization problem using the below Eq. (1).

$\begin{aligned} \gamma^{*}, \emptyset^{*}, \lambda^{*}=\operatorname{argmin}_{(\gamma, 0, \lambda)} D(q(\theta, Z, \beta / \gamma, \phi, \lambda) / /& \\ p(\theta, z, \beta / D ; \alpha, \eta) & \end{aligned}$  (1)

where, γ, ϕ and λ denotes the free variational parameters we surmised θ, z and β with, separately. Here D(q||p) denotes the KL divergence among q and p. What's more, by changing γ, ϕ, and λ, we get distinctive q appropriations having various distances from the genuine posterior p. We will likely discover the γ*, ϕ*, and λ* that limit the KL divergence between the surmised q and the genuine posterior p.

Obfuscation: Data obfuscation is a form of data veiling where information is intentionally mixed/supplanted with sham information to prevent unauthorized access to sensitive materials [18]. This form of scrambling/replacement results in unintelligible or confusing data. Data obfuscation at this point scrambles the data in a manner that retains readability. All the keywords which are extracted in the previous step would be replaced with keywords in the custom-built dictionary by applying the "FlashText algorithm". Figure 3 shows an example of obfuscating the record-2, in which original information is replaced with dummy information, which is colored in yellow.

2.3.png

Figure 3. Data obfuscation example

4.png

Figure 4. Process of determining n, d, and e values

Cryptography: Is a process of keeping an information secret. i.e. it transforms the information into a form that can't be read and understand by other users who are not authorized. It has four concerns, confidentiality, integrity, non-repudiation, and authentication. In our work, we use cryptography techniques to ensure the confidentiality information stored in the cloud.

At this level, the obfuscated document will be encrypted by using a public key encryption algorithm RSA, which can be implemented practically and yields better results than the rest of the algorithms. The working procedure of RSA is given below.

The encryption and decryption process are done using Eq. (2) and Eq. (4) respectively.

$C=M^{\wedge} e \bmod n$  (2)

$M=C^{\wedge} d \bmod n$  (3)

where, C and M are the ciphertext and Plaintext.

Here the value of 'n', should be known to the cloud service provider as well as the cloud service consumer. And service provider knows the value of 'e’ and only authorized consumer knows the value of ‘d’. Here public key is {e , n } and private key is {d , n }.

And, to determine the values of e , d , n we have to follow the steps below.

Step-1: Choose two numbers p and q which are huge primes.

Step-2: Calculate ‘n’ as the multiplication of  p and q.

Step-3: Choose some large integer randomly, d , such that $G C D(d,((p-1) *(q-1)))=1$

Step-4: Find the value of ‘e’ such that $e * d=1(\bmod ((p-1) *(q-1)))$

Based on these values and following the RSA procedure, the input gets converted to ciphertext. The process of determining n, d, e values is shown in Figure 4.