A plagiarism-resistant iterative truth value discovery method based on lightweight homomorphic encryption

Abstract

The present invention relates to the technical field of crowdsourcing systems, and specifically to an anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption. By adding the attribute of the working end weight in the truth value calculation process of the crowdsourcing task, the truth value is inferred according to the attribute of the working end weight to obtain a preliminary set of truth value answers, and whether the working end is the original author is determined based on the preliminary set of truth value answers. Based on the answer set of the original author, the plagiarist is determined by the similarity threshold, and the weight of the plagiarist is reduced. The next round of truth value calculation is performed according to the changed weight of the working end until the convergence criterion is met to output the final set of truth value answers. The crowdsourcing system can automatically monitor and automatically exclude plagiarism in the process of truth value calculation, thereby improving the accuracy of the final truth value result.

Description

Technical Field

[0001] The present invention relates to the technical field of crowdsourcing systems, and in particular to an anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption. Background Art

[0002] Yin et al. formally defined the truth discovery problem for the first time and proposed the TruthFinder model. The model uses Bayesian distribution to iteratively evaluate the weights and truth values ​​of data sources. Since then, a large number of truth discovery methods have been proposed. For example, Li et al. proposed the CRH model, which uses an optimization model to handle heterogeneous data. The truth discovery technology has subsequently been widely cited in the crowdsourcing field to obtain accurate estimated truth values.

[0003] With the rapid development of crowdsourcing platforms, people are paying more and more attention to privacy issues. Li et al. used the parameters provided by the workers to generate noise samples based on exponential distribution, and then added this unique noise to the data to form Gaussian distribution noise, thereby completing the data aggregation process; Pang et al. improved personalized classification based on privacy protection, and achieved high-precision truth discovery while protecting privacy by adjusting the privacy level; Sun et al. considered providing personalized rewards based on the privacy needs of workers to meet the privacy preferences of different workers; Tang et al. solved the user privacy problem through anonymization methods and proposed a perturbation method to prevent data from being obtained by malicious third parties. The traditional encryption mode homomorphic encryption is introduced into the field of truth discovery. This method does not modify the worker's answer, but directly converts the worker's answer from the plaintext domain to the ciphertext domain, and then performs truth aggregation and weight update according to different homomorphic encryption algorithms.

[0004] Some studies have considered source correlation in crowdsourcing. Dong et al. proposed a method to consider source correlation in the truth discovery scenario, arguing that there may be dependencies between sources that always provide the same wrong answers; Jiang et al. comprehensively considered the duplication and incentive mechanism in crowdsourcing, and the detected duplicators may not receive rewards; Wang et al. focused on duplicators and multi-valued objects and proposed a comprehensive Bayesian method with fine-grained duplication detection; Li et al. proposed a duplication detection method that considers the scale of big data and the number of sources. Chen et al. proposed the CONAN method, which aims to maximize the use of the duplicator's answers to enhance aggregation accuracy. However, none of these works consider the case where the data is encrypted or perturbed. That is, existing technologies include differential privacy, data anonymization, and homomorphic encryption. Although homomorphic encryption can be used for truth discovery, this encryption method often ignores plagiarism between workers. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption, comprising the steps of:

[0006] S1. The key distributor generates a key pair through a homomorphic encryption algorithm, sends the public key in the key to the working end, sends the private key to the data requesting end through an encrypted channel, and executes the next step.

[0007] S2. The crowdsourcing server sends the crowdsourcing task in the form of multiple-choice questions proposed by the data requesting end to at least one working end. Each working end solves the working task to establish an answer set, encrypts the answer set with a public key, and uploads it to the crowdsourcing server to execute the next step.

[0008] S3. The crowdsourcing server performs a cyclic truth calculation process on all the encrypted answer sets to obtain a true answer set. The truth calculation process includes three sub-processes: truth inference, replication detection and working end weight update. The three sub-processes of truth inference, replication detection and working end weight update are performed in parallel in a cycle. When multiple rounds of the truth calculation process meet the preset convergence criteria, the next step is executed.

[0009] S4. Send the truth value answer set obtained by the truth value calculation to the data request end, and the data request end decrypts the truth value answer set according to the private key.

[0010] Preferably, the truth inference subprocess obtains a predicted set of truth answers based on the weight of the working end and the number of selected answers.

[0011] Preferably, the copy detection sub-process includes determining that the work end is the original author based on the preset weight of the work end and when the degree of similarity between the answer set of the corresponding work end and the true answer set reaches a correct threshold, traversing the answer sets of other work ends based on the answer set of the original author, and when the degree of similarity between the answer sets of other work ends and the answer set of the original author reaches a set similarity threshold, determining that the work end that exceeds the similarity threshold is a potential duplicator.

[0012] Preferably, the preset weight of the working end is initialized each time the working task is executed.

[0013] Preferably, the preset weight of the working end is the weight when the last work task was completed.

[0014] Preferably, the working end weight update sub-process includes lowering the set weight value of the potential duplicator when the working end is determined to be a potential duplicator, determining the potential duplicator to be a true duplicator when the weight of the potential duplicator is reduced to a set recognition threshold, and increasing the set weight value of the corresponding working end when the similarity between the answer set of the working end and the true value answer set exceeds a set correct threshold.

[0015] Preferably, when the working end is determined to be a true replicator, the answer set of the true replicator is deleted during the truth value inference process.

[0016] Preferably, the convergence criterion is that the set of truth answers for the current truth inference is the same as the set of truth answers for the previous truth inference.

[0017] Preferably, the convergence criterion is that the true value calculation cycle is performed twenty times.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. Add the attribute of work-end weight in the truth value calculation process of the crowdsourcing task, perform truth value inference based on the attribute of work-end weight to obtain a preliminary set of truth answers, and determine whether the work-end answer set is the answer set of the original author based on the preliminary set of truth answers. Based on the answer set of the original author, traverse the answer sets of other work ends. When the similarity between the answer sets of other work ends and the answer set of the original author exceeds the similarity threshold, determine that the work end that exceeds the similarity threshold is a potential copier, and reduce the weight of the potential copier until the weight of the work end is lower than the recognition threshold and then exclude it from subsequent truth value calculations. Perform the next round of truth value calculation based on the changed weight of the work end until the convergence criterion is met to output the final set of truth answers. Enable the crowdsourcing system to automatically monitor and automatically exclude plagiarism during the truth value calculation process, thereby improving the accuracy of the final truth value result;

[0020] 2. The truth value calculation process is performed in an encrypted state, which improves the accuracy of the truth value calculation while maintaining the privacy of the data.

[0021] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the process of the present invention;

[0023] Figure 2 A functional structural diagram of a crowdsourcing system according to the present invention; DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] See also Figure 1-Figure 2 The present invention provides an anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption, comprising the steps of:

[0026] S1. The crowdsourcing system usually includes a key distributor, a working end, a data requesting end, and a crowdsourcing server, wherein the working end and the data requesting end are both electrically connected to the crowdsourcing server and the key distributor, and the crowdsourcing server is electrically connected to the key distributor. First, a key pair is generated by the key distributor through a homomorphic encryption algorithm. In this embodiment, the homomorphic encryption algorithm is the Paillier homomorphic encryption algorithm. The public key in the key is sent to the working end, and the private key is sent to the data requesting end through an encrypted channel, and the next step is executed.

[0027] S2. The crowdsourcing server sends crowdsourcing tasks in the form of multiple-choice questions proposed by the data request end, such as disease identification, dog breed classification, and sentiment analysis, to at least one working end. The multiple-choice questions in the crowdsourcing task include at least one multiple-choice question and the corresponding multiple-choice question includes at least one answer option. Each of the working ends answers the work task, that is, selects an answer option for the multiple-choice question, and establishes an answer set for all the multiple-choice questions and their corresponding answer options. After completing the crowdsourcing task, the working end uses the public key to encrypt the answer set. The encrypted answer set will form an answer set of the multiple-choice question and its corresponding answer ciphertext. The encrypted answer set is uploaded to the crowdsourcing server to execute the next step.

[0028] S3. The crowdsourcing server performs a cyclic truth value calculation process on all the encrypted answer sets to obtain a true value answer set. The truth value calculation process includes three sub-processes: truth value inference, copy detection, and work end weight update. The truth value inference, copy detection, and work end weight update are performed in parallel in three sub-processes. The truth value inference sub-process obtains a predicted true value answer set according to the weight of the work end and the weight of the selected answer. The copy detection sub-process includes determining that the work end is the original author based on the preset weight of the work end and the similarity between the answer set of the corresponding work end and the true value answer set when the correct threshold is reached, and traversing the answer sets of other work ends based on the answer set of the original author. Based on the characteristics of the Paillier homomorphic encryption algorithm, when the similarity between the answer sets of other work ends and the answer set of the original author reaches the set similarity threshold, the work end exceeding the similarity threshold is determined to be a potential copier. The work end weight update sub-process includes reducing the set weight value of the potential copier when the work end is determined to be a potential copier, and determining that the potential copier is a true copier when the weight of the potential copier is reduced to the set recognition threshold. Based on the characteristic of the Paillier homomorphic encryption algorithm that can realize homomorphic addition operations, the algorithm for judging similarity can be implemented by using conventional existing similarity algorithms such as Jaccard similarity coefficient, Hamming distance, and cosine similarity. In this embodiment, the judgment of similarity uses a similarity matrix to judge the similarity of the answer set in the encrypted state. The principle of using the similarity matrix to judge the similarity of the answer sets of two working ends in this embodiment is as follows: Assuming that there are two users, RA and RB, and the tasks are TA, TB, and TC, then there are two similarity matrices, namely MA and MB. Assuming that A did TA, TB, and B did TB, TC, then MA= ,MB= , where all D are ciphertext. It can be seen that RA and RB can only be replicated on TB. Assuming DBB=DAB, then MA= ,MB= , if DBB is not equal to DAB, then MA= ,MB= , then you only need to determine whether their answers on TB are the true values ​​of this round. If so, it is determined that both working terminals have chosen the correct answer and cannot be considered plagiarism. If not, both working terminals have chosen the wrong answer and there is a possibility of plagiarism. Relying on this possibility of the wrong answer being regarded as a copy and combining the weights of each working terminal, the respective copying situations can be obtained. When the similarity between the answer set of the working terminal and the true value answer set exceeds the set correct threshold, the set weight value of the corresponding working terminal is increased. That is, in the first round of truth value calculation, the true value answer set is first obtained according to the weight of the working terminal and the number of selected answers. When the preset weight of the weight of the working terminal is the initial weight, the weights of all working terminals are the same, and the answer with the largest number of selected answers will be obtained as the true value answer in the first round of truth value calculation. Then, based on the first round of true value answer set, the answer sets of all working terminals are traversed. When the content similarity between the answer set of the working terminal and the true value answer set of the first round reaches the correct threshold, the working terminal is determined to be the original author. The setting of the correct threshold is independently set according to the actual situation, such as 70%; or it can be set separately according to the number of work terminals and the number of multiple-choice questions in the crowdsourcing task. For example, when the number of multiple-choice questions is less than fifty, the correct threshold is set to 80%, and when the number of multiple-choice questions exceeds fifty, the correct threshold is set to 60%.

[0029] When the original author is determined, the answer set of the original author is used as a benchmark to traverse the answer sets of all other working terminals for comparison, and the potential copiers are determined by the similarity threshold. The similarity threshold is also set according to the actual situation, such as 90% or 100%. The weight of the working terminal that exceeds the similarity threshold is reduced. After the first round of truth value calculation process is completed, the weight of the working terminal will change, and then the second round of truth value calculation will be performed. At this time, due to the different weights of the working terminals, the second round of true value answer set obtained in the second round of truth value calculation process no longer simply considers the number of answers selected, but also considers the weight of the working terminal. Therefore, the second round of true value answer set may be different from the first round of true value answer set. At this time, based on the second round of true value answer set, the answer sets of all working terminals are traversed, and the original author is judged again based on the set correct threshold, and then the copier is judged based on the original author's answer set. After reducing the weight of the copier, the third truth value calculation process is performed based on the weights obtained by all working terminals in the second round of truth value calculation. The truth value calculation is repeated until the convergence criterion is reached.

[0030] In this embodiment, the convergence criterion is set to the set of truth answers for the current truth inference to be the same as the set of truth answers for the previous truth inference. Since the number of working terminals in the crowdsourcing system is variable and unpredictable, in order to prevent tens of thousands of working terminals from solving crowdsourcing tasks at the same time, which results in a long time for truth calculation, the convergence criterion is further set to the truth calculation cycle to be performed twenty times.

[0031] There are two ways to set the preset weights of the work end: one is to initialize it after each work task is completed, and the weight changes of the work end are limited to the current work task; the other is that the weight changes of the work end are retained and continued to the subsequent work tasks. You can choose one of the two ways to set the preset weights or a combination of both. For example, when the form of the multiple-choice question is sentiment analysis, the preset weight of the subsequent sentiment analysis type of work tasks is set to the last weight in the previous work task, so that the weight changes of the same type of work tasks can be continued; when the subsequent multiple-choice question form is another type of work task that is completely different from sentiment analysis, such as disease identification, the weight initialization of the work end is set, so that the weight of the work end of the same type of work task can be continued, and the weights of the work ends of different types of work tasks are independent of each other, and similar settings.

[0032] Furthermore, when the weight of the working end identified as a potential copier is further reduced to the set recognition threshold, the potential copier is determined to be a true copier, and the answer set of the true copier is deleted in the subsequent truth value calculation process to exclude the answer set of the true copier. The set recognition threshold is also set manually according to the specific situation. For example, the initial weight is 1. When the answer set of the working end is determined to exceed the similarity threshold with the answer set of the original author, the weight is reduced by 0.1, and the recognition threshold is set to 0.5. When the weight of the working end is reduced to 0.5 after five cycles in the truth value calculation process, the content in the answer set of the working end is no longer considered in the sixth round of truth value calculation. By continuously updating the weight of the working end, the answer set of the true copier who is a pure plagiarist is eventually excluded, and a true value answer set that is closer to the truth is obtained. When multiple rounds of truth value calculation process meet the preset convergence criteria, the next step is executed;

[0033] S4. Send the truth value answer set obtained by the truth value calculation to the data request end, and the data request end decrypts the truth value answer set according to the private key. The working end can only encrypt, and decryption can only be performed by the data request end that holds the private key, so that the privacy of the truth value answer set is effectively guaranteed.

[0034] On the basis of the above implementation scheme, the processors, modules, corresponding control programs, algorithm programs and other supporting technologies mentioned in the present invention can be implemented in combination with existing electrical technology, information technology, software technology and general protocols. They are not within the scope of protection required by the present invention and will not be described in detail in this application.

[0035] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered as exemplary and non-restrictive in all respects.

Claims

1. A plagiarism-resistant iterative truth value discovery method based on lightweight homomorphic encryption, characterized in that: Includes steps: S1. The key distributor generates a key pair through a homomorphic encryption algorithm, sends the public key in the key to the working end, sends the private key to the data requesting end through an encrypted channel, and executes the next step; S2. The crowdsourcing server sends the crowdsourcing task in the form of a multiple-choice question proposed by the data requesting end to at least one working end, each of which solves the work task to establish an answer set, encrypts the answer set using a public key and uploads it to the crowdsourcing server to execute the next step; S3. The crowdsourcing server performs a cyclic truth value calculation process on all the encrypted answer sets to obtain a truth value answer set. The truth value calculation process includes three sub-processes: truth value inference, replication detection, and working end weight update. The three sub-processes of truth value inference, replication detection, and working end weight update are performed in parallel. When multiple rounds of truth value calculation processes meet the preset convergence criteria, the next step is executed; S4. Sending the truth value answer set obtained by the truth value calculation to the data request end, and the data request end decrypts the truth value answer set according to the private key; In the step S2, the truth inference obtains a predicted set of truth answers according to the weight of the working end and the number of selected answers; The duplication detection includes determining that the working terminal is the original author based on the preset weight of the working terminal and when the degree of similarity between the answer set of the corresponding working terminal and the true answer set reaches a correct threshold, and traversing the answer sets of other working terminals based on the answer set of the original author. When the degree of similarity between the answer sets of other working terminals and the answer set of the original author reaches a set similarity threshold, determining that the working terminal that exceeds the similarity threshold is a potential duplicator.

2. According to claim 1, the anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption is characterized in that: The preset weight of the working end is initialized each time the working task is executed.

3. The anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption according to claim 1 is characterized in that: The preset weight of the working end is the weight when the last working task was completed.

4. The anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption according to claim 1 is characterized in that: The working end weight update includes lowering the set weight value of the potential duplicator when the working end is judged to be a potential duplicator, judging the potential duplicator to be a real duplicator when the weight of the potential duplicator is reduced to a set recognition threshold, and increasing the set weight value of the corresponding working end when the similarity between the answer set of the working end and the true answer set exceeds a set correct threshold.

5. The anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption according to claim 2 is characterized in that: When the working end is determined to be a true replica, the answer set of the true replica is deleted during the truth value inference process.

6. The anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption according to claim 1 is characterized in that: The convergence criterion is that the set of truth answers for the current truth inference is the same as the set of truth answers for the previous truth inference.

7. The anti-plagiarism iterative truth value discovery method based on lightweight homomorphic encryption according to claim 1 is characterized in that: The convergence criterion is that the true value calculation cycle is performed twenty times.

Images