Block chain-based bidirectional evaluation trusted data perception method

Abstract

The invention discloses a two-way evaluation trusted data perception method based on a block chain. The method aims to solve the problem of low data quality caused by hostile attacks in the crowd sensing network. Firstly, potential safety hazards caused by centralized attacks such as DDOS on a current centralized platform are avoided by using a block chain technology; meanwhile, the truth value discovery algorithm is executed by recruiting edge computing nodes to obtain truth values to evaluate the credibility of the data participants, and the influence of malicious data pinched by malicious participants on the overall data quality is recognized and reduced. And the service quality of the data participant is obtained through the reputation obtained by evaluating the friendliness, the timeliness, the cooperation degree and other capabilities of the data requester when the data participant completes the task and the data collection capability of the reputation. The data requester recruits the data participants under limited budget by combining the service quality of the data participants with the reverse task auction, and maximizes the effectiveness of the data requester to obtain the data so as to construct a high-quality application.

Description

Technical Field

[0001] The present invention belongs to the field of credible data collection in a crowd intelligence network, and in particular to a method for quickly and at low cost obtaining the credibility of participants in a crowd intelligence network and collecting real data. Background Art

[0002] The crowd-sensing network is a new type of participatory perception data collection paradigm. In the traditional crowd-sensing network, the data requester publishes the data collection task information through the central platform, including the data location, content and time, and recruits data participants to complete the task by paying rewards to achieve the purpose of collecting data to construct applications. In the crowd-sensing network, data participants can often be people or sensors with smart bracelets, smart phones or other sensor devices. They can use the sensor devices to collect data to obtain corresponding rewards. The traditional crowd-sensing network is based on a centralized platform core, which is easily affected by centralized DDOS attacks. Once the central platform is breached by the attacker, the entire crowd-sensing network will collapse, and various information will be directly exposed to the attacker. Therefore, the traditional crowd-sensing network has great security risks.

[0003] At the same time, the quality of the collected data directly affects the quality of the application constructed by the data requester, and is an important part of the crowd-sensing network. However, the data collected directly from data participants is uneven, and there are also malicious participants in the network who hope to get rewards or submit false, forged malicious data in order to attack the system. These malicious data will reduce the quality of the collected data, and then reduce the quality of the application constructed by the data requester. Therefore, it is of great significance to identify data participants, distinguish between trusted participants and malicious participants, and select trusted participants in the recruitment process to collect high-quality data and construct high-quality applications.

[0004] The most important thing in data participant identification is how to identify the data submitted by the data participants. When the data submitted by the data participants is within a certain error range from the real data, the data of the data participants is often valid, while the data with too large an error or fabricated out of thin air is not only invalid, but also affects the true value discovery process of other valid data. However, how to identify whether this data is valid is a difficult problem. First of all, the data collected by crowd sensing is often time-space related. Since the environment when the data participants collect data is often not exactly the same, there will inevitably be certain errors. If the true value cannot be obtained, it is impossible to identify valid and invalid data. For example, the noise data at the same location is often different due to different collection locations, times, and equipment. In addition, there are a lot of data in the crowd sensing network that cannot be obtained multiple times, which makes verification more difficult.

[0005] Based on the above problems, it is necessary to conduct a truth discovery algorithm. For the data of a certain task, when the majority of the participants are trusted, since the data they collect is independent and identically distributed, the truth value obtained by executing the truth discovery algorithm on the data is often a reliable truth value and close to the data submitted by the reliable participants, while the forged malicious data often deviates from the truth value, thereby achieving the identification of the data. In addition, we can exclude untrustworthy participants by recruiting participants based on trust, so that the majority of the participants recruited to complete the data collection task are trustworthy. However, for the data requester, the effectiveness of obtaining data is not only related to the data perception ability of the data participant, but also to multiple dimensions such as the attitude, timeliness and collaboration when completing the task, and the task bidding price. Therefore, it is not enough to recruit data participants based on trust alone.

[0006] Based on the above challenges, the method of the present invention first replaces the centralized platform with blockchain to prevent security issues such as DDOS, and at the same time evaluates data participants from two dimensions, namely, trust related to data perception ability and reputation related to attitude, timeliness, collaboration and other related aspects when completing tasks, and selects by integrating the service quality and price of the two dimensions to maximize the data utility obtained by the data requester. At the same time, for the trust of data participants, we recruit edge nodes to execute the truth discovery algorithm, evaluate the data submitted by the participants through the true value, and update the reputation of the participants through the evaluation score of the data requester. In addition, we introduce trust to prevent malicious scoring by data requesters, and complete the two-way evaluation update of data requesters and data participants through their scoring. Summary of the invention

[0007] The present invention discloses a two-way evaluation trusted data perception method based on blockchain. The invention method is to solve the problem of low data quality caused by malicious attacks in the crowd intelligence perception network. First, by using blockchain technology, the current centralized platform is protected from security risks caused by centralized attacks such as DDOS. At the same time, by recruiting edge computing nodes to execute the truth value discovery algorithm to obtain the truth value to evaluate the trust of data participants, identify and reduce the impact of malicious data fabricated by malicious participants on the overall data quality. And the reputation obtained by evaluating the attitude, timeliness, collaboration and other abilities of data participants when completing tasks by the data requester is combined with their data collection capabilities to obtain the service quality of the data participants. The data requester recruits data participants under a limited budget by combining the service quality of the data participants with the reverse task auction, and maximizes the utility of the data requester to obtain data to build high-quality applications.

[0008] The technical solution of the invention is as follows:

[0009] (1) The blockchain is mainly composed of data requesters, data participants and edge computing nodes. In this method, data requesters obtain perception data by issuing data collection tasks. Data publishing tasks are periodic, with rounds as the basic unit of data collection. t = {1, 2, 3…i…n} represents the number of cycles of data collection. Data participants obtain rewards by applying for data collection tasks and submitting perception data. Data requesters obtain true values ​​by recruiting edge computing nodes to execute truth discovery algorithms. Among them, the service quality of data participants consists of two parts. On the one hand, it is the trust of data quality related to data, and on the other hand, it is the reputation of non-data related capabilities such as collaboration, timeliness, and friendliness during the data collection process. The evaluation of trust comes from the quality evaluation of the data submitted by the data requester, and the evaluation of reputation comes from the evaluation of the data requester who performs the task. At the same time, the evaluation of the data requester on the data participant is not necessarily true. In order to prevent malicious evaluation by the data requester, the trust of the data requester is evaluated. Specifically, the error between the true value obtained by the data requester and the real data is minimized through the following steps;

[0010] (2) The set of data requesters is R = {r1, r2..., r i ,...,r |R|}, publish data collection tasks through blockchain, data participants W = {w1,w2,…,w j ,…,w |W|}By viewing the task list, a request for a task is initiated and a quotation is made for the task. The jth data participant w j The quote is b j The i-th data requester r i The trust y of each data participant, such as the jth data participant j and reputation j The initial value is set to 0.5 and is continuously updated during the subsequent data collection process; the data requester r i After receiving the task application from the data participants, K data participants are selected based on their trust, reputation and quotation:

[0011] a. First, through each data participant w j Trust level j and reputation j Calculate the quality of service of data participants

[0012] q j =y j ·v j .

[0013] b. Secondly, through the task bidding of each data participant j To calculate the participants' service quality-to-cost ratio

[0014]

[0015] c. The data requester sorts the data participants according to the service quality cost ratio and selects the top K task participants to perform the data perception task, and after performing the task, the data participant w j Give Reward

[0016]

[0017] in, and They respectively represent the service quality cost ratio and task bidding of the K+1th data participant after sorting.

[0018] (3) After the data participants submit the perceived data, the data requester recruits edge computing nodes to execute the truth discovery algorithm, aggregates the data of the data participants, and obtains the true value. The specific process is as follows: For the data perception task T of the data requester i , the recruited data participants w j The data submitted is The edge computing node obtains the true value by iteratively calculating the true value and estimating the weight:

[0019] True value calculation:

[0020]

[0021] Weight estimation:

[0022]

[0023] in, and They represent the estimated true value and data participant w at the tth iteration respectively. j The weight of the recruited edge nodes is calculated iteratively until the true value error between the tth round and the t-1th round is less than a given relatively small threshold ε, for example, the value of ε can be 0.001, that is,

[0024]

[0025] (4) The mean of the Beta distribution B(α,β) is used to estimate the trust of the data requester, the trust and reputation of the data participant. The mean of the Beta distribution can be expressed as

[0026]

[0027] Update the parameters α and β to update the mean

[0028]

[0029] Where ρ is the update step size, ω t is the weight of the update, the default value is 1, e -μσ is the exponential decay used for updates, σ is the decay factor that changes over time, and m t Determines whether the update mean is increasing or decreasing.

[0030] a. Reputation update of data participants: After receiving the data and the true value, the data requester will evaluate the data participant based on the degree of satisfaction. i For data participants w j The rating is When the score is greater than the threshold γ, we increase the reputation of the data participant, otherwise we reduce its reputation; through the above update method and set

[0031] ω t =x i .

[0032]

[0033] Among them, x i Indicates that the requester r i The trust level, γ is the threshold, which is a hyperparameter and can be taken as 0.5;

[0034] b. Data participant trust update: When data participant w j Submit task T i Data After that, the data quality is evaluated by the following formula:

[0035]

[0036] in, Represents task T i The iterative true value of When it is greater than the threshold δ, we increase the trust of the data participant, otherwise we reduce his trust. Update the distribution through data quality

[0037]

[0038] c. Update of data requester’s trustworthiness: For data requester, the requester r i For data participants w j The rating is The data participant w j The reputation is v j ,When the data requester’s score deviates from the reputation of the data participant, i.e. , it means that its rating is wrong, and its trust is reduced, where θ is a hyperparameter threshold value that can be taken as 0.05, otherwise its trust is increased.

[0039]

[0040] Beneficial Effects

[0041] The present invention discloses a two-way evaluation trusted data perception method based on blockchain. Compared with previous methods, the invention has the following beneficial effects: First, in previous methods, the data utility collected by task data participants is only related to the data perception ability of the data participants, that is, the trust, while ignoring the influence of the attitude, timeliness and collaboration of the data participants when completing the task on the final collected data. This method evaluates the reputation of data participants to integrate these factors, and derives their final service quality based on their data perception ability, so as to better identify data participants with high service quality. At the same time, this method is also combined with the reverse auction method, by calculating the service quality cost ratio of each data participant and comprehensively considering the cost, so that the data requester can maximize the data utility under a limited budget. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of the mechanism of the method of the present invention;

[0043] Figure 2 To compare the task completion rate of previous methods;

[0044] Figure 3 A comparison chart of the total data utility of different methods under different budgets for data requesters DETAILED DESCRIPTION

[0045] In order to facilitate the understanding of the present invention, the present invention will be described in more comprehensive and detailed manner below in conjunction with the accompanying drawings and preferred embodiments of the specification, but the protection scope of the present invention is not limited to the following specific embodiments.

[0046] Unless otherwise defined, all professional terms used below have the same meanings as those generally understood by those skilled in the art. The professional terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the scope of protection of the present invention.

[0047] Unless otherwise specified, various raw materials, reagents, instruments and equipment used in the present invention can be purchased from the market or prepared by existing methods.

[0048] Example:

[0049] In the construction of smart cities, we often need to collect real-time data about the environment in the city, such as traffic flow, noise, temperature, PM2.5 and other data at various locations in the city. At this time, citizens carrying smart collection and smart bracelets in the city and driving vehicles are often data participants, who can collect and perceive the environmental data in real time. Data requesters can often obtain these real-time data through certain rewards. The experimental method is as follows: 1,000 data participants are set, and there are 560 perception tasks. The trust and reputation of the participants are randomly generated. The data submitted by the data participants are generated according to their trust and reputation. The trust and reputation make the data errors submitted by the participants always smaller than the data errors of participants with low trust and reputation. In addition, participants with low trust and reputation are malicious participants, who will carry out poisoning attacks by randomly fabricating and false data.

[0050] The experimental results of the inventive method are given below.

[0051] Figure 2 The given are the completion rates of participants using different methods. One experimental method is the completion rate of the method of the present invention, and the other is the task completion rate without considering the reputation dimension of the participants' attitude, timeliness, and collaboration in completing the task. From the experimental results, it can be seen that the task completion rate using the method of this paper is much higher than the previous method that only considers trust.

[0052] Figure 3 It is the total data utility of different algorithms considering the trust and reputation of participants under different budgets of data requesters. The difference between the total data utility of the optimal algorithm (assuming that the platform knows the service quality of the data participants and recommends the data participants with the highest service quality in each round) and the total data utility of other algorithms is called regret. We implemented several other algorithms for comparison with the method of the present invention. The comparison algorithms include a selection method based on an upper bound of confidence, a greedy exploration method with a probability of 0.3 (randomly selecting data participants with a probability of 0.3, and selecting data participants based on the evaluated service quality with a probability of 0.7), and a random selection method (randomly selecting data participants). It can be found that the method of the present invention has the highest total data utility when other methods also consider the reputation dimension, and the regret remains basically unchanged when the regret of other methods increases linearly with the increase of the budget.

Claims

1. A two-way evaluation trusted data perception method based on blockchain, which mainly includes the following steps: (1) The blockchain is mainly composed of data requesters, data participants, and edge computing nodes. In this method, data requesters obtain perception data by issuing data collection tasks. The data publishing tasks are periodic, with rounds as the basic unit of data collection. t = {1, 2, 3…i…n} represents the number of cycles of data collection. Data participants obtain rewards by applying for data collection tasks and submitting perception data. Data requesters obtain truth values ​​by recruiting edge computing nodes to execute truth discovery algorithms. The service quality of data participants consists of two parts: on the one hand, the trust of data quality related to data, and on the other hand, the reputation of non-data related capabilities such as collaboration, timeliness, and friendliness during data collection. The evaluation of trust comes from the quality evaluation of the data submitted, and the evaluation of reputation comes from the evaluation of the data requester who performs the task. At the same time, the evaluation of the data requester on the data participant is not necessarily true. In order to prevent malicious evaluation by the data requester, the trust of the data requester is evaluated. Specifically, the error between the true value obtained by the data requester and the real data is minimized through the following steps: (2) The set of data requesters is R = {r1, r2..., r i ,...,r |E| }, publish data collection tasks through blockchain, data participants W = {w1,w2,…,w j ,…,w |w| }By viewing the task list, a request for a task is initiated and a quotation is made for the task. The jth data participant w j The quote is b j The i-th data requester r i ,,The trust y of each data participant, such as the jth data participant j and reputation j The initial value is set to 0.5 and is continuously updated during the subsequent data collection process; the data requester r i After receiving the task application from the data participants, K data participants are selected based on their trust, reputation and quotation: a. First, through each data participant w j Trust level j and reputation j Calculate the quality of service of data participants q j =y j ·v j . b. Secondly, through the task bidding of each data participant j To calculate the participants' service quality-to-cost ratio c. The data requester sorts the data participants according to the service quality cost ratio and selects the top K task participants to perform the data perception task, and after performing the task, the data participant w j Give Reward in, and They respectively represent the service quality cost ratio and task bidding of the K+1th data participant after sorting. (3) After the data participants submit the perceived data, the data requester recruits edge computing nodes to execute the truth discovery algorithm, aggregates the data of the data participants, and obtains the true value. The specific process is as follows: For the data perception task T of the data requester i ,,Recruited data participants w j The data submitted is The edge computing node obtains the true value by iteratively calculating the true value and estimating the weight: True value calculation: Weight estimation: in, and They represent the estimated true value and data participant w at the tth iteration respectively. j The weight of the recruited edge nodes is iterated until the true value error between the tth round and the t,,-1th round is less than a given relatively small threshold ε, for example, the value of ε can be 0.001, that is, (4) The mean of the Beta distribution β(α,β) is used to estimate the trust of the data requester, the trust and reputation of the data participant. The mean of the Beta distribution can be expressed as Update the parameters α and β to update the mean Where ρ is the update step size, ω t is the weight of the update, the default value is 1, e -μσ is the exponential decay used for updates, σ is the decay factor that changes over time, and m t Determines whether the update mean is increasing or decreasing. a. Reputation update of data participants: After receiving the data and the true value, the data requester will evaluate the data participant based on the degree of satisfaction. i For data participants w j Rated V j i ,When the score is greater than the threshold γ, we increase the reputation of the data participant, otherwise reduce its reputation; through the above update method and set oh t =x i . in, x i Indicates that the requester r i The trust level, γ is the threshold, which is a hyperparameter and can be taken as 0.5; b. Data participant trust update: When data participant w j Submit task T i Data After that, the data quality is evaluated by the following formula: in, Represents task T i The iterative true value of When it is greater than the threshold δ, we increase the trust of the data participant, otherwise we reduce his trust. Update the distribution through data quality c. Update of data requester’s trustworthiness: For data requester, the requester r i For data participants w j Rated V j i , and data participant w j The reputation is v j ,When the score of the data requester deviates from the reputation of the data participant, that is, |V j i -v j When |≥θ, it means that the rating is wrong and its trust is reduced. θ is a hyperparameter threshold value that can be taken as 0.

05. Otherwise, its trust is increased.

Images