Source and truth value credibility estimation method and system in multi-data source environment

Abstract

The invention belongs to the technical field of data mining and anomaly detection, and particularly relates to a source and truth value credibility estimation method and system in a multi-data-source environment, and the method comprises the steps: 1, carrying out the conversion and discretization of data, and generating data; step 2, performing parameter estimation based on an EM algorithm; on the basis of truth value estimation, the posterior probability occurring in the interval where the value is located is approximate to the credibility of the truth value; and step 3, estimating the credibility of the data source based on anomaly detection. According to the method, firstly, an observed value matrix is converted into a declaration value matrix based on truth value estimation, then, an EM algorithm is applied to estimate core parameters, truth value credibility calculation is carried out on estimated values of the parameters, and the credibility of the parameters is estimated by calculating the posterior probability of truth value estimation. And finally, based on truth value estimation and the observed value matrix, calculating the credibility of each data source by adopting an anomaly detection method.

Description

Technical Field

[0001] This invention belongs to the field of data mining and anomaly detection technology, specifically relating to a method and system for estimating the credibility of sources and truth values ​​in a multi-data source environment. Background Technology

[0002] With the rapid development of technologies such as the Internet of Things, big data, and artificial intelligence, data is being generated at an unprecedented rate. However, finding the desired information from this vast and complex data is becoming increasingly difficult. This is not only due to errors in the information or slow data updates, but also because different data sources may provide conflicting descriptions of the same entity. Therefore, researching how to discover true information in a multi-data source environment is essential; this is the truth-finding problem.

[0003] Currently, there is a wealth of research on the truth-finding problem in multi-source data. Most of this research is based on information from different data sources, comprehensively considering factors such as the credibility of the data sources, prior knowledge, and data quality, and using iterative methods to jointly deduce the truth. Therefore, the credibility of the data source plays a crucial role in the truth-finding problem. For example, the classic Truth Finder algorithm iteratively calculates the truth value of the data and the credibility of the data source; the higher the credibility of the data source, the more reliable the data it provides; furthermore, the more high-credibility data a data source provides, the higher its credibility.

[0004] For the problem of data source credibility estimation, in addition to methods that combine truth derivation with data analysis, there are also some methods specifically designed for data source credibility estimation. For example, a solution specifically for web data source credibility estimation is proposed. This method relies on the correctness of the factual information provided by the data source. By using joint inference in a new multi-level probabilistic model, it distinguishes between errors generated during the extraction process and actual errors in the data source itself, thus reliably calculating the true credibility of the data source. Another example is a global replication detection algorithm that considers the case of data source replication. This algorithm determines the credibility of the data source by detecting replication relationships between data sources.

[0005] In the truth-finding problem, in addition to the reliability of the data source, the reliability of the truth value estimated by the algorithm is also very important. However, there are not many papers specifically studying the problem of truth reliability estimation. Most papers use labels or scores to reflect the performance of truth-finding algorithms and use this performance to represent the reliability of the truth. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention proposes a method and system for estimating the credibility of truth values ​​and data sources in a multi-data source environment. The core idea of ​​this invention is to abstract observations from different data sources into different events and apply the EM algorithm to estimate core parameters. Based on these core parameters, the credibility of truth values ​​and data sources can be estimated.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for estimating source and truth credibility in a multi-data source environment, comprising the following steps:

[0009] Step 1. Data transformation and discretization to generate parameters.

[0010] Step 2. Perform parameter estimation based on the EM algorithm; and based on the true value estimation, use the posterior probability of the interval in which the value occurs to approximate the confidence level of the true value.

[0011] Step 3. Estimate the credibility of the data source based on anomaly detection.

[0012] Furthermore, step 1 specifically includes the following steps:

[0013] Step 1.1: Timestamp: This invention divides a certain time period T into N equally spaced time segments, and uses a timestamp T. n This represents the nth time segment.

[0014] Step 1.2: Data Source: A data source is a device or raw medium that provides a certain required data, such as a sensor. This invention assumes there are a total of D data sources S1, S2, ..., S... D .

[0015] Step 1.3: Truth Value: Truth value refers to the actual numerical value of a certain attribute of a real-world entity, such as a person's real age or the temperature of a body of water at a specific moment. This invention only considers the case of a single truth value, that is, a certain attribute of an entity has one and only one truth value.

[0016] Step 1.4: Observations: Time-series data obtained by continuously observing entities from the data source. For example, a water temperature sensor monitors a body of water for T = 1 minute, obtaining N = 60 time-series data points of water temperature. Data Source S d At timestamp T n The obtained observations are represented by y nd This means that the matrix Y = (y1, y2, ..., y3) is defined by combining the observations from all data sources at all timestamps. N ), where y n For timestamp T n The vector of observed values ​​within.

[0017] Step 1.5: Declared Value: After calculating the estimated true value y*, the declared value is calculated based on the difference between the observed value and the estimated true value. (Using x...) nd ∈{0,1} represents the data source S d At timestamp T n The declared value on. Specifically, if |y nd -y*|<ε (ε>0 is a very small adjustable parameter), then let x nd =1, otherwise let x =1 nd =0. All data sources are at the same timestamp T. n The declared values ​​within are combined into a vector x. n =(x n1 x n2 , ..., x nD ) T Furthermore, the declared values ​​of all timestamps are combined to define a matrix X = (x1, x2, ..., x...). N ).

[0018] Step 1.6: Events: Discretize the numerical attributes of the observed target and label them with different events. This invention assumes that the normal attribute value of the observed target is within the range (a, b), for example, the normal range of water temperature can be set to (-10, 40). Divide this range into K equally spaced sub-intervals. Specifically, if the observed value is within sub-interval B... k If the value is within the range of ((k-1)(ba) / K+a, k(ba) / K+a), then define event e. k (k = 1, 2, ..., K) occurred. Here, K is an adjustable parameter.

[0019] Step 1.7: Based on the above definition of events, assume π k (k = 1, 2, ..., K) represents event e. k The prior probability of occurrence.

[0020] Therefore, we can obtain

[0021]

[0022] Step 1.8: Further, define μ kd For event e k Under the conditions that occur, data source S d The conditional probability with a declared value of 1. Assume μ kd It does not change with time. Let the parameter θ = (π, U), where π = (π1, π2, ..., π). K ) T U = (μ1, μ2, ..., μ) K ) Tμ k =(μ k1 μ k2 , ..., μ kD ) T , k∈{1,2,...,K}.

[0023] Furthermore, step 2 specifically includes the following steps:

[0024] Step 2.1: Calculate the initial parameter estimate θ 0 =(π) 0 U 0 For ease of description, we define [N] = {1, 2, ..., N} and [D] = {1, 2, ..., D}.

[0025] Step 2.2: Considering that the EM algorithm is sensitive to initial values, empirical probabilities are used to calculate the initial parameters. Specifically, let...

[0026]

[0027]

[0028] Where, B = {y nd |y nd ∈B k , n∈[N]}

[0029] Step 2.3: EM Algorithm:

[0030] Step 2.3.1: E-step: Let θ i To obtain the parameter estimates for the i-th iteration, the Q-function needs to be computed in the E-step of the (i+1)-th iteration:

[0031]

[0032] Where Z = [z nk ] is a latent variable; P(Z|X,θ) i ) is the conditional probability distribution of latent variables given the observed data and the current parameter estimates.

[0033] Step 2.3.2: Define the latent variable Z = (z1, z2, ..., z...) N ), where z n Indicates at timestamp T n Within, the occurrence of each event: If event e k If it occurs, then z n0 , z n1 , ..., z nK There is only z in the middle nk The value is 1, and the rest are 0.

[0034] Step 2.3.3: Finally, we can obtain...

[0035]

[0036]

[0037] in,

[0038]

[0039] It is important to note here that when θ i Given, γ(z) nk It can be calculated from , so it can be regarded as a constant.

[0040] Step 2.3.4: M-step: Based on the Q-function form obtained from the E-step calculation, the Q-function can be made to adjust for μ. kd and π k Find the partial derivatives for each parameter and then find the parameter value that makes the partial derivative zero. This yields the parameter μ. k Iterative formula

[0041]

[0042] in,

[0043] Step 2.3.5: In E[z] nk Given the given conditions, reconsider the M-step optimization problem:

[0044]

[0045]

[0046] parameter π k The update formula is

[0047]

[0048] Step 2.4: Based on parameter μ k , π k Update the formula, define the number of algorithm iterations i. If the number of iterations is less than i, repeat step 2.3, iterating until the specified number of iterations i is reached or the parameters converge, then return the parameter estimate.

[0049] Step 2.5: Based on parameter estimation Calculate the confidence level of the true estimate y*, i.e., P(e j =1|X), therefore, the confidence level of the true value estimate y* is

[0050]

[0051] Note that here the confidence level of the true value estimate y* is approximated by the interval B in which its value lies. j The posterior probability of occurrence.

[0052] Furthermore, step 3 specifically includes the following steps:

[0053] Step 3.1: Assume Y d ={y 1d y 2d , ..., y Nd} is the data source S d The set of observation data within N timestamps, where d∈[D].

[0054] Step 3.2: Considering Y d There may be abnormal data in the data; calculate Y directly. d The sample standard deviation will also take these outliers into account, which will affect the estimation quality of the sample standard deviation. Therefore, in order to reduce the impact of outliers on the sample standard deviation, an iterative method is used here to calculate Y. d The sample standard deviation.

[0055] Step 3.3: Calculate Y d The initial sample standard deviation, i.e.

[0056]

[0057] Step 3.4: Based on the current estimate of the sample standard deviation σ i-1 Identify outlier data, i.e., calculate the set.

[0058]

[0059] Step 3.5: Remove outlier data from the observed data, and then calculate the sample standard deviation of the remaining data, i.e.

[0060]

[0061] Step 3.6: Let i = i + 1. Repeat steps 3.4 and 3.5 until no more outliers are found. Finally, output the estimated sample standard deviation.

[0062] Step 3.7: After obtaining the data source S d After estimating the sample standard deviation, the set of normal data can then be calculated.

[0063]

[0064] Step 3.8: Given a normal data y nd ∈C dIts credibility is defined as

[0065]

[0066] Here, erf() is the Gaussian error function.

[0067] Step 3.9: In step 3.8, first use the Gaussian error function to convert the original observed data y nd The probability of ∈[0,+∞) is mapped to "abnormal", and then its confidence level s is calculated. nd ∈[0,1). Finally, based on the reliability of all normal data, the data source S can be calculated. d Credibility

[0068]

[0069] As can be seen from the above formula, the credibility of the data source is the "average" credibility of all the normal data it outputs. Here, the average is calculated using the amount of data N output by the data source.

[0070] This invention also discloses a system based on the above-mentioned source and truth credibility estimation method in a multi-data source environment, which includes the following modules:

[0071] Data production module: performs data conversion and discretization, and produces parameters;

[0072] Truth discovery module: performs parameter estimation based on the EM algorithm; and approximates the credibility of the truth value by using the posterior probability of the interval in which the value is located based on the truth value estimation; and estimates the credibility of the data source based on anomaly detection.

[0073] This invention discloses a method and system for estimating source and truth credibility in a multi-data source environment. First, the invention converts the observation matrix into a declaration matrix based on truth estimation. Then, the EM algorithm is applied to estimate the core parameters, and the credibility of these estimated parameters is calculated by determining their posterior probabilities. Finally, based on the truth estimation and the observation matrix, anomaly detection methods are used to calculate the credibility of each data source. Attached Figure Description

[0074] Figure 1 This is a flowchart of a source and truth credibility estimation method in a multi-data source environment according to an embodiment of the present invention;

[0075] Figure 2 This is a comparison chart of the actual number of anomalies in each data source according to the embodiments of the present invention and the number of anomalies calculated by the method proposed in the present invention;

[0076] Figure 3This is a comparison chart of the final credibility of each data source calculated according to an embodiment of the present invention with the actual credibility of each data source.

[0077] Figure 4 This is a graph showing the changes in truth value reliability error and data source reliability error as the proportion of abnormal data increases, according to an embodiment of the present invention.

[0078] Figure 5 This is a block diagram of a source and truth credibility estimation system under a multi-source data environment according to an embodiment of the present invention. Detailed Implementation

[0079] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0080] This embodiment presents a method for estimating the source and truth credibility in a multi-data source environment, such as... Figure 1 As shown, the specific steps are as follows:

[0081] Step 1. Perform data transformation and discretization, production parameters, related parameters and definitions.

[0082] Step 1.1: In this embodiment, the number of events K = 8, the number of data sources D = 4, each data source contains N = 1000 timestamps of data, and their size range is within the interval (0,8). The number of EM algorithm iterations i is 15.

[0083] Step 1.2: For each data source S d Assume that its normal observed values ​​follow an expected value of y = 5 and a variance of . Gaussian distribution, Independently and uniformly sample from the interval (1,2). Assume the truth value is estimated by the truth discovery module as y* = 4.5.

[0084] Step 1.3: To ensure the reliability of normal data, it is restricted to within two expected standard deviations, meaning each normal data point satisfies...

[0085] Step 1.4: For each data source S d The method for generating abnormal data is as follows: First, randomly generate w d There are 10 data points, all of which are outside the expected two standard deviations; then, w are randomly selected from the original normal data. d At each of the following locations, normal data is replaced with abnormal data. Finally, the same process is performed on each data source (w1=20, w2=30, w3=50, w4=80) to obtain four time series data, each containing a different number of abnormal data.

[0086] Step 2. Building upon Step 1, a truth credibility estimation method based on the EM algorithm is proposed. Details are as follows:

[0087] Step 2.1: Calculate the initial parameter estimate θ 0 =(π) 0 U 0 The initial values ​​for this parameter are initialized for iteration, as shown in Table 1.

[0088] Table 1

[0089]

[0090] Step 2.2: After 15 iterations of the EM algorithm, the parameter estimates are obtained. As shown in Table 2:

[0091] Table 2

[0092]

[0093] Step 2.3: Next, calculate the confidence level of the true value based on the parameter estimation results. In Step 1.1, the entire observation interval is divided into K equal intervals, where B k =((k-1)(ba) / K+a,k(ba) / K+a) is the kth subinterval. Therefore, substituting a=0, b=8, and K=8, we can get y*=4.5∈B5.

[0094] Step 2.4: The confidence level of the true value estimate y* is approximately the interval B in which its value lies. j Given the posterior probability of occurrence, the confidence level of the true estimate y* is:

[0095]

[0096] Step 3. Building upon Step 1, a data source credibility estimation method based on anomaly detection is proposed. Details are as follows:

[0097] Step 3.1: Based on the estimated true value y* and the observation matrix Y, the source credibility estimation module calculates the credibility of each data source based on the anomaly detection method.

[0098] Step 3.2: Transfer data source S d The number of abnormal data detected is denoted as PN. d Then PN d =N-|C d |, where C d A collection of normal data for each data source.

[0099] Step 3.3: In this embodiment, the number of abnormal data points detected by the method proposed in this invention are PN1=24, PN2=34, PN3=52 and PN4=79, respectively. Figure 2 As shown, by comparing the actual number of abnormal data points contained in the data source, it can be demonstrated that the method proposed in this invention has a high accuracy rate.

[0100] Step 3.4: In this embodiment, the confidence levels of each data source obtained using the method proposed in this invention are s1 = 0.951, s2 = 0.935, s3 = 0.901, and s4 = 0.875, respectively. Figure 3 As shown, by comparing the authenticity and reliability of the data source, it can be demonstrated that the method proposed in this invention has a high accuracy rate.

[0101] Step 3.5: To better verify the accuracy of the method proposed in this invention, experimental simulations were conducted on the proportion of abnormal data from different data sources. Based on the credibility obtained from the simulation and the true credibility, the true credibility error was calculated.

[0102]

[0103] and data source credibility error

[0104]

[0105] Experimental results are as follows Figure 4 As shown, the results demonstrate that the method proposed in this invention can effectively estimate the credibility of the truth value and the credibility of the data source.

[0106] like Figure 5 As shown in the figure, this embodiment discloses a source and truth credibility estimation system in a multi-source data environment, which includes the following modules:

[0107] Data production module: performs data conversion and discretization, and produces parameters;

[0108] Truth discovery module: performs parameter estimation based on the EM algorithm; and approximates the credibility of the truth value by using the posterior probability of the interval in which the value is located based on the truth value estimation; and estimates the credibility of the data source based on anomaly detection.

[0109] Other aspects of this embodiment can be found in the embodiments described above.

[0110] The above embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

Claims

1. A method for estimating source and truth credibility in a multi-data source environment, characterized in that, Includes the following steps: Step 1: Perform data transformation and discretization to generate parameters; Step 2: Parameter estimation is performed based on the EM algorithm; and based on the true value estimation, the posterior probability of the interval in which the value occurs is approximated as the confidence level of the true value. Step 3: Estimate the credibility of the data source based on anomaly detection.

2. The source and truth reliability estimation method in a multi-data source environment according to claim 1, characterized in that, Step 1 specifically includes the following steps: Step 1.1: Timestamp: Divide a time period T into N equally spaced time segments and use timestamps T to represent them. n This represents the nth time segment; Step 1.2: Data Sources: Assume there are a total of D data sources S1, S2, ..., S... D ; Step 1.3: Truth value: Only consider the case of a single truth value, that is, a certain attribute of an entity has one and only one truth value; Step 1.4: Observations: Data Source S d At timestamp T n The obtained observations are represented by y nd This means that the matrix Y = (y1, y2, ..., y3) is defined by combining the observations from all data sources at all timestamps. N ), where y n For timestamp T n The vector of observed values ​​within; Step 1.5: Declared Value: After calculating the estimated true value y*, the declared value is calculated based on the difference between the observed value and the estimated true value; using x... nd ∈{0,1} represents the data source S d At timestamp T n The declared value on; put all data sources at the same timestamp T n The declared values ​​within are combined into a vector x. n =(x n1 x n2 , ..., x nD ) T Combine all timestamp declaration values ​​to define a matrix X = (x1, x2, ..., x...). N ); Step 1.6: Events: Discretize the numerical attributes of the observed target and label them with different events; assume that the normal attribute value of the observed target is within the range (a, b); divide this range into K equally spaced sub-intervals; if the observed value is in sub-interval B... k If the value is within the range of ((k-1)(ba) / K+a, k(ba) / K+a), then define event e. k It happened, k = 1, 2, ..., K; where K is an adjustable parameter; a is the minimum value among the normal observations, and b is the maximum value among the normal observations; Step 1.7: Based on the definition of an event, assume π k For event e k The prior probability of occurrence; therefore, we obtain Step 1.8: Define μ kd For event e k Under the conditions that occur, data source S d The conditional probability with a declared value of 1; assume μ kd It does not change over time; let the parameter θ = (π, U), where π = (π1, π2, ..., π). K ) T U = (μ1, μ2, ..., μ) K ) T μ k =(μ k1 μ k2 , ..., μ kD ) T , k∈{1,2,...,K}.

3. The source and truth reliability estimation method in a multi-data source environment according to claim 2, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Calculate the initial parameter estimate θ 0 =(π) 0 U 0 Define [N] = {1, 2, ..., N} and [D] = {1, 2, ..., D}; Step 2.2: Calculate the initial parameters; let Where, B = {y nd |y nd ∈B k ,n∈[N]}; Step 2.3: EM Algorithm Iteration: Step 2.3.1: E-step: Let θ i Given the parameter estimates for the i-th iteration, compute the Q-function in the E-step of the (i+1)-th iteration: Where Z = [z nk ] is a latent variable; P(Z|X,θ) i () is the conditional probability distribution of latent variables given the observed data and the current parameter estimates; Step 2.3.2: Define the latent variable Z = (z1, z2, ..., z...) N ), where z n Indicates at timestamp T n Within, the occurrence of each event: If event e k If it occurs, then z n0 , z n1 , ..., z nK In the middle, z nk The value is 1, and all others are 0; Step 2.3.3: Obtain in, When θ i Given, γ(z) nk It can be calculated and considered as a constant; Step 2.3.4: M-step: Based on the Q-function form obtained from the E-step calculation, let the Q-function be related to μ. kd and π k Find the partial derivatives for each parameter and then find the parameter value that makes the partial derivative zero; thus, obtain the parameter μ. k Iterative formula in, Step 2.3.5: In E[z] nk Given the given conditions, reconsider the M-step optimization problem: parameter π k The update formula is Step 2.4: Based on parameter μ k , π k Update the formula, define the number of algorithm iterations i. If the number of iterations is less than i, repeat step 2.3, iterating until the specified number of iterations i is reached or the parameters converge, then return the parameter estimate. Step 2.5: Based on parameter estimation Calculate the confidence level of the true estimate y*, i.e., P(e j =1|X), therefore, the confidence level of the true estimate y* is 4. The source and truth reliability estimation method in a multi-data source environment according to claim 3, characterized in that, Step 3 specifically includes the following steps: Step 3.1: Assume Y d ={y 1d y 2d , ..., y Nd } is the data source S d A set of observation data within N timestamps, where d∈[D]; Step 3.2: Calculate Y using an iterative method. d The sample standard deviation; Step 3.3: Calculate Y d The initial sample standard deviation, i.e. Step 3.4: Based on the current estimate of the sample standard deviation σ i-1 Identify outlier data, i.e., calculate the set. Step 3.5: Remove outlier data from the observed data, and then calculate the sample standard deviation of the remaining data, i.e. Step 3.6: Let i = i + 1; repeat steps 3.4 and 3.5 until no outliers are found; finally, output the estimated sample standard deviation. Step 3.7: After obtaining the data source S d After estimating the sample standard deviation, calculate the set of normal data. Step 3.8: Given a normal data y nd ∈C d Its credibility is defined as Where erf() is the Gaussian error function; Step 3.9: In step 3.8, first use the Gaussian error function to convert the original observed data y nd The probability of anomaly is mapped to ∈[0,+∞), and then its confidence level s is calculated. nd ∈[0,1); Finally, based on the confidence level of all normal data, the data source S is calculated. d Credibility 5. A source and truth reliability estimation system in a multi-source environment, characterized in that, Includes the following modules: Data production module: performs data conversion and discretization, and produces parameters; Truth discovery module: performs parameter estimation based on the EM algorithm; and approximates the credibility of the truth value by using the posterior probability of the interval in which the value is located based on the truth value estimation; and estimates the credibility of the data source based on anomaly detection.

6. The source and truth reliability estimation system in a multi-data source environment according to claim 5, characterized in that, The data production module is as follows: definition: Timestamp: Divide a time period T into N equally spaced time segments and use timestamps T to represent them. n This represents the nth time segment; Data sources: Assume there are a total of D data sources S1, S2, ..., S... D ; Truth value: Only considers the case of a single truth value, that is, a certain attribute of an entity has one and only one truth value; Observations: Data source S d At timestamp T n The obtained observations are represented by y nd This means that the matrix Y = (y1, y2, ..., y3) is defined by combining the observations from all data sources at all timestamps. N ), where y n For timestamp T n The vector of observed values ​​within; Declaration value: After calculating the estimated true value y*, the declaration value is calculated based on the difference between the observed value and the estimated true value; expressed as x. nd ∈{0,1} represents the data source S d At timestamp T n The declared value on; put all data sources at the same timestamp T n The declared values ​​within are combined into a vector x. n =(x n1 x n2 , ..., x nD ) T Combine all timestamp declaration values ​​to define a matrix X = (x1, x2, ..., x...). N ); Event: Discretize the numerical attributes of the observed target and label them with different events; assume that the normal attribute value of the observed target is within the range (a, b); divide this range into K equally spaced sub-intervals; if the observed value is in sub-interval B... k If the value is within the range of ((k-1)(ba) / K+a, k(ba) / K+a), then define event e. k It happened, k = 1, 2, ..., K; where K is an adjustable parameter; a is the minimum value among the normal observations, and b is the maximum value among the normal observations; Based on the definition of an event, assume π k For event e k The prior probability of occurrence; therefore, we obtain Define μ kd For event e k Under the conditions that occur, data source S d The conditional probability with a declared value of 1; assume μ kd It does not change over time; let the parameter θ = (π, U), where π = (π1, π2, ..., π). K ) T U = (μ1, μ2, ..., μ) K ) T μ k =(μ k1 μ k2 , ..., μ kD ) T , k∈{1,2,...,K}.

7. The source and truth reliability estimation system in a multi-data source environment according to claim 6, characterized in that, In the truth discovery module, parameter estimation is performed based on the EM algorithm; and based on the truth estimation, the posterior probability of the interval in which the value occurs is used to approximate the confidence level of the truth, as follows: Calculate the initial parameter estimate θ 0 =(π) 0 U 0 Define [N] = {1, 2, ..., N} and [D] = {1, 2, ..., D}; Calculate the initial parameters; let Where, B = {y nd |y nd ∈B k ,n∈[N]}; EM Algorithm Iteration: Step E: Remember θ i Given the parameter estimates for the i-th iteration, compute the Q-function in the E-step of the (i+1)-th iteration: Where Z = [z nk ] is a latent variable; P(Z|X,θ) i () is the conditional probability distribution of latent variables given the observed data and the current parameter estimates; Define the latent variable Z = (z1, z2, ..., zn) N ), where z n Indicates at timestamp T n Within, the occurrence of each event: If event e k If it occurs, then z n0 , z n1 , ..., z nK In the middle, z nk The value is 1, and all others are 0; get in, When θ i Given, γ(z) nk It can be calculated and considered as a constant; M-step: The form of the Q-function obtained based on the E-step calculation. Let the Q function be relative to μ kd and π k Find the partial derivatives for each parameter and then find the parameter value that makes the partial derivative zero; thus, obtain the parameter μ. k Iterative formula in, In E[z nk Given the given conditions, reconsider the M-step optimization problem: parameter π k The update formula is Based on parameter μ k , π k The update formula defines the number of algorithm iterations, i. If the number of iterations is less than i, the EM algorithm is repeated until the specified number of iterations i is reached or the parameters converge, at which point the parameter estimate is returned. Based on parameter estimation Calculate the confidence level of the true estimate y*, i.e., P(e j =1|X), therefore, the confidence level of the true estimate y* is 8. The source and truth reliability estimation system in a multi-data source environment according to claim 7, characterized in that, In the truth discovery module, the confidence level of the data source is estimated based on anomaly detection as follows: Assume Y d ={y 1d y 2d , ..., y Nd } is the data source S d A set of observation data within N timestamps, where d∈[D]; Y is calculated using an iterative method. d The sample standard deviation; Calculate Y d The initial sample standard deviation, i.e. Based on the current estimate of the sample standard deviation σ i-1 Identify outlier data, i.e., calculate the set. Outlier data is removed from the observed data, and then the sample standard deviation of the remaining data is calculated. Let i = i + 1; repeatedly calculate the sample standard deviation of the set and the remaining data until no outliers are found; finally, output the estimated sample standard deviation. After obtaining data source S d After estimating the sample standard deviation, calculate the set of normal data. Given a normal data y nd ∈C d Its credibility is defined as Where erf() is the Gaussian error function; In this process, the Gaussian error function is first used to convert the original observed data y nd The probability of anomaly is mapped to ∈[0,+∞), and then its confidence level s is calculated. nd ∈[0,1); Finally, based on the confidence level of all normal data, the data source S is calculated. d Credibility

Images