A multi-truth discovery method based on the reliability of data source domain
By constructing explicit/implicit observation maps in the fully connected field and reliability modeling of data source fields, combining the professionalism of data source fields and the trustworthiness scores of multi-true observations, the professionalism of data source fields in the existing technology is solved. The ignored problem realizes automatic truth recognition and reasoning in a multi-truth environment, improving decision-making efficiency and quality.
Patent Information
- Application Number
- CN202211005282.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-22
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-08-22
AI Technical Summary
The existing multi-truth discovery method affects the algorithm performance when assuming that the distribution of hidden variables does not match the actual data, and ignores the field expertise of the data source, resulting in the inability to obtain the reliability of the fine data source, affecting decision efficiency and quality.
Through reliability modeling in the data source field, a fully connected field explicit/implicit observation map is constructed, and the data source field professionalism, object field similarity and multi-true observation trust scores are used to balance the data distribution long-tail phenomenon , calculate the domain reliability of the data source and the true value probability of the observed value, and infer the complete true value on the object properties.
It realizes automatic identification and inference of correct and complete truth values in the data conflict environment where multiple truth values exist, improves decision efficiency and quality, provides more refined data source reliability estimates, and enhances multi-truth discovery. Accuracy and completeness.
Smart Images

Figure CN115392018B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a multiple truth value discovery method, in particular to a multiple truth value discovery method based on data source field reliability, belonging to the technical field of data integration. Background Art
[0002] With the rapid development of the Internet, various social networks and crowdsourcing platforms are generating Web data every moment. Behind the massive Web data lies huge commercial and research value. However, the reliability of websites or platforms that provide Web data varies greatly, resulting in data conflicts in the observation values provided by multiple data sources for the same object attributes. Truth discovery has emerged as a key technology to solve the data conflict problem. At present, multi-truth discovery methods have been successfully applied to many scenarios, such as big data fusion systems, the construction of high-quality knowledge bases, crowd perception, privacy protection, crowdsourcing aggregation, and medical fields.
[0003] In data conflicts, some object attributes have unique truth values, such as the opening price of a stock on a certain day, and some object attributes have multiple truth values, such as the author of a book. When there are multiple truth value data conflicts in object attributes, finding all the truth values from several multi-truth attribute values provided by the data source about the object is called multi-truth discovery. As one of the important technologies for resolving data conflicts and improving the quality of Web data, the truth value discovery method has been widely valued by researchers.
[0004] Bo Zhao and others from the University of Illinois at Urbana-Champaign first proposed a probabilistic graphical method LTM to achieve multiple truth discovery. They assumed that the recall and specificity of the data source follow the Beta distribution to construct a probabilistic graphical model and use a sampling-based method to infer multiple truth values (Zhao B, Rubinstein BIP, Gemmell J, et al. A bayesian approach to discovering truth from conflicting sources for data integration [J]. arXiv preprint arXiv, 2012.1203.0058: p.550-561.). Subsequently, the research group proposed a probabilistic graphical model GTM to solve the data conflict problem of continuous observations by modeling the relationship between the reliability of the data source and the truth value and the observation value (Zhao B, Han JA probabilistic model for estimating re-al-valued truth from conflicting sources [J]. Proc. of QDB, 2012, 1817.). The above two methods start from the two dimensions of data source quality (recall and specificity) and can solve most of the multiple truth data conflict problems. However, when the distribution of the hidden variables assumed by the method does not match the actual data, the performance of the algorithm will be greatly affected.
[0005] Xianzhi Wang et al. from the University of Adelaide proposed a Bayesian method for multi-truth discovery, which takes into account the mapping relationship between data sources and observations, and combines the mutual exclusion relationship between observations and data source duplication detection to achieve multi-truth discovery (Wang X, Sheng QZ, Fang XS, et al. An integrated bayesian approach for effective multi-truth discovery[C] / / Proceedings of the 24th ACM Internationalon Conference on Information and Knowledge Management. 2015: p.493-502.). Xueling LIN et al. from the Hong Kong University of Science and Technology and Ma Ruxia et al. from Renmin University of China both believe that observations provided by the same data source in different fields should be assigned different weights. Xueling LIN et al. from the Hong Kong University of Science and Technology combined the observation credibility score with the domain expertise of the data source and proposed a DART model that integrates the Bayesian method (Lin X, Chen L. Domain-aware multi-truth discovery from conflicting sources [J]. Proceedings of the VLDB Endowment, 2018, 11(5): p.635-647.). Ma Ruxia et al. from Renmin University of China proposed the CTruthFinder algorithm, which uses the Bayesian method to iteratively calculate the reliability of data source classification and the credibility of observations (Ma Ruxia, Meng Xiaofeng. Research on truth discovery method based on data source classification credibility [J]. Computer Research and Development. 2015, 52(09): p.1931-1940.). Ma Ruxia et al. from Renmin University of China proposed a Mtruths multi-truth discovery method, which transforms the multi-truth data conflict problem into an optimization problem, and obtains multiple truth values by enumeration and greedy algorithm to find the optimal solution. Lu Jing et al. from Shanghai University of Technology classified data sources according to the long-tail phenomenon of data distribution, and used attribute set correlation to seek the optimal distribution of data source weights to obtain multiple truth values (Lu Jing, Hu Cheng, Liu Cong. Research on multi-truth discovery method using attribute set correlation and source error [J]. Small Microcomputer Systems. 2019, 40(03): pp. 601-605.). Fang et al. from Macquarie University in Sydney, New South Wales, Australia proposed a SmartVote method to quantify the relationship between two types of data sources to accurately estimate the reliability of the data source. In addition, it also combines the popularity, mutual exclusion and looseness of the object, and the long-tail phenomenon of the data source to achieve more accurate multi-truth discovery.The above methods all ignore the prior knowledge of the domain expertise of the data source and fail to obtain the precise reliability of the data source, which may lead to performance loss. Summary of the invention
[0006] In response to the above problems, the present invention provides a multi-truth value discovery method based on the reliability of data source domain, which can automatically identify, judge, and infer the complete truth value in a data conflict environment where there are multiple truth values of an attribute, thereby helping users make decisions and improving decision-making efficiency and quality.
[0007] To achieve the above purpose, the multi-truth value discovery method based on the reliability of data source domain specifically includes the following steps:
[0008] Step 1: Data source domain reliability modeling;
[0009] Step 1-1: Calculate the domain richness of the data source in For data sources i In the field j Provides data richness within the domain by calculating the domain d j Internal data sources i The percentage of the number of objects provided to the total number of objects is obtained The calculation formula is as follows:
[0010]
[0011] Where N(s i , d j ) indicates the data source s i In the field j The number of objects provided in Indicates that in the field d j A collection of data sources that provide object description information;
[0012] Step 1-2: Calculate the data source s i In the field j Professionalism within
[0013] Step 1-3: Calculate the data source s i With object o k Domain similarity I(o k ,s i );
[0014] Step 1-4: Construction of fully connected domain explicit / implicit observation graph, in domain d j Data source collection In the object collection The provided explicit / implicit observation relationship is modeled as a fully connected domain explicit / implicit observation graph Data Source For the picture The vertices in
[0015] Step 1-5: Graph the explicit / implicit observations in the domain In, calculate the data source s i With data sources m Provides consensus of explicit / implicit observations of objects
[0016] Step 1-6: Graph the explicit / implicit observations in the domain In, calculate the data source s i For data sources m Long-tail compensation on explicit / implicit observations and
[0017] Step 1-7: Explicit / implicit observations in the domain map Inside, calculate the vertex s i Point to vertex s m Edge weight
[0018] Step 1-8: Normalization calculation of inter-vertex weights, calculation of inter-vertex normalized weights and
[0019] Step 1-9: Calculate the reliability of the data source domain, in domain d j Explicit / implicit observation value graph of internal fully connected domain Perform random walks to obtain the data source s in the explicit / implicit observation value graph i The stationary access probability p(s i )and Calculation data sources i In the field j The accuracy within and false positive rate Get data sources i In the field j Reliability within the domain;
[0020] Step 2: Observation credibility modeling;
[0021] Calculate the probability of the observation v being true / false based on the domain reliability of the data source and the relationship between the observations provided by the data source
[0022] Step 3: True value inference;
[0023] when When , the observed value v is judged to be true, otherwise it is judged to be false, completing the true value inference.
[0024] Calculate the data source s in step 1-2 i In the field j Professionalism within The calculation formula is as follows:
[0025]
[0026] Where α is a hyperparameter for adjusting the weight of domain expertise. In the experiments on the dataset, α is set to 1.5.
[0027] Calculate the data source s in steps 1-3 i With object o k Domain similarity I(o k ,s i ), all data sources s1~s i Corresponding areas d1~d j Professional composition matrix The formula is as follows:
[0028]
[0029] According to the object k The domain construction object o k The field vector The formula is as follows:
[0030]
[0031] In the formula, when the object o k Belongs to field d j When otherwise
[0032] Calculation data sources i With object o k Domain similarity I(o k ,s i )The formula is as follows:
[0033]
[0034] In the formula For data sources i The domain vector of The i-th row vector of .
[0035] In steps 1-5, in the domain explicit / implicit observation value diagram In, data sources i With data sourcesm Provides consensus of explicit / implicit observations of objects The calculation formula is as follows:
[0036]
[0037]
[0038] In the formula and Respectively represent data sources i and m In the field j a collection of objects providing observations, and Respectively represent data sources i and m In object o k The set of explicit observations provided, and Respectively represent data sources i and m In object o k The set of implicit observations provided, σ(v) and They represent the probability that the observed value v is true / false, respectively. and Indicates data source i With data sources m In object o k provides the similarity of explicit / implicit observations on and The calculation formula is as follows:
[0039]
[0040]
[0041] μ(s i , o k )and Respectively represent data sources i In object o k The credibility score of the explicit observation provided on the attribute, data source s m In object o k The implicit observation value provided by the attribute is a credibility score, and the explicit / implicit observation value provided by the multi-truth data conflict data source is assigned a credibility score μ(s i , o k )and The calculation formula is as follows:
[0042]
[0043]
[0044] Where U(o k ) represents the object o k The complete set of observations.
[0045] Graph of explicit / implicit observations in the domain in steps 1-6 In, data sources i For data sources m Long-tail compensation on explicit / implicit observations and The calculation formula is as follows:
[0046]
[0047]
[0048] In the formula I(o k ,s m ) indicates the data source s m With object o k The domain similarity of , σ(v) and They represent the probability that the observed value v is true / false, μ(s m , o k )and Respectively represent data sources m In object o k The credibility score of the explicit observation provided on the attribute, data source s m In object o k Provides a confidence score for the implicit observation on the attribute.
[0049] Calculate the vertex s in steps 1-7 i Point to vertex s m Edge weight The calculation formula is as follows:
[0050]
[0051]
[0052] Where λ is the damping factor, ensuring the domain explicit / implicit observation value map has a stationary distribution, Indicates data source i With data sources m In the field j Provides the consensus of explicit / implicit observations, Indicates data source i For data sources m In the field j The long-tail compensation on explicit / implicit observations is provided.
[0053] Normalized weights between vertices in steps 1-8 and The calculation formula is as follows:
[0054]
[0055]
[0056] In the formula Represents the explicit / implicit observation value graph in the domain Midpoints i Point to vertex s m The weight of the edge.
[0057] Accuracy in steps 1-9 and false positive rate The calculation formula is as follows:
[0058]
[0059]
[0060] In the formula and are the pre-set parameters. Represents the data source s in the explicit / implicit observation value diagram i The stationary access probability.
[0061] In step 2, the probability of the observation v being true / false is calculated based on the domain reliability of the data source and the relationship between the observations provided by the data source. The calculation formula is as follows:
[0062]
[0063]
[0064]
[0065]
[0066] In the formula Indicates data source i In the field j The accuracy / false positive rate within Indicates data source i In object o k The final accuracy / false positive rate in the field, and represents the set of data sources that provide explicit / implicit observations v, ″o k Represents object ok A collection of fields. Indicates that the object o is provided k The data source collection of observations.
[0067] Compared with the existing technology, this multi-truth value discovery method based on the domain reliability of the data source models the explicit / implicit observation value relationship provided by the data source as a domain explicit / implicit observation value graph, combines the domain expertise of the data source, the similarity between the data source and the object domain, and the credibility score of the multi-truth value observation value, and balances the loss caused by the long tail phenomenon of data distribution to achieve a more precise data source reliability estimation, thereby inferring the correct and complete truth value of the object attribute. In a data conflict environment where there are multiple truth values of the attribute, it can automatically identify, judge, and infer the complete truth value, thereby helping users, enterprises and governments make decisions and improving decision-making efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] Figure 1 This is a flow chart of the multi-truth value discovery method based on the reliability of the data source domain. DETAILED DESCRIPTION
[0069] This multi-truth value discovery method based on the domain reliability of the data source introduces the prior knowledge of the domain expertise of the data source. It does not need to assume hidden variables in advance. By constructing a domain explicit / implicit observation value graph, combining the domain similarity between the data source and the object, and the credibility score of the multi-truth observation value, the domain reliability of the data source is obtained by balancing the loss caused by the long tail phenomenon of the data distribution, and the correct and complete truth value of the object attribute is inferred. Specifically: first calculate the domain expertise of the data source; secondly calculate the domain similarity between the data source and the object, initialize the parameters and the credibility of the observation value, and then use the iterative calculation method to jointly derive the domain reliability of the data source and the credibility of the observation value. In each iteration, the multi-truth value reasoning method based on the domain reliability of the data source is used to infer the data domain source reliability according to the credibility of the observation value, the domain expertise of the data source, the domain similarity between the data source and the object, and the credibility score of the multi-truth observation value, and then re-infer the credibility of the observation value through the domain reliability of the data source. When the performance index of the algorithm changes within a certain threshold range, it is considered that the algorithm has converged; finally, all the truth values of the object attributes are inferred. The flowchart of this multi-truth value discovery method based on the domain reliability of the data source is as follows: Figure 1 shown.
[0070] In order to facilitate those skilled in the art to understand the embodiment of the present invention, the following uses three data sources providing author information of a book as an example to explain relevant technical terms in the embodiment of the present invention.
[0071] Object: represents an identifiable and unique thing in the real world. For example, in Table 1 below, a book is an object.
[0072] Attribute: used to describe the characteristics of a specific object. For example, in Table 1 below, the author attribute of a book.
[0073] Data source: indicates the source of data for collecting the attribute description information of related objects. For example, in Table 1 below, a book website collects information about book authors, and a book website is a data source.
[0074] Explicit-observation: It means that a data source directly provides descriptive information about an object attribute. For example, in Table 1 below, the names of the authors of books directly provided by the data source are explicit observations.
[0075] Observation set: The set of explicit observations of an object provided by all data sources is called the observation set of the object. For example, the observation set of the author attribute of a book in Table 1 below is {Erdemci, Fulya; Kocak, Orhan; Germaner, Semra; Stompbox, Paul}.
[0076] Implicit observation: The complement of the set of explicit observations provided by the data source relative to the full set of object observations is called the implicit observation provided by the data source. For example, Table 1 below shows the implicit observations provided by the data source on the book.
[0077] Domain: Indicates the scope to which an object belongs. For example, "art" is a domain, and a book can belong to the "art" domain.
[0078] Table 1 Three data sources provide author information for a book
[0079]
[0080] A method for discovering multiple truth values based on the reliability of data source domains includes the following steps:
[0081] Step 1: Data source domain reliability modeling
[0082] Step 1-1: Introduce the richness of data source domain Calculate. Among them, For data sources i In the field j Provides data richness within the domain by calculating the domain d j Internal data sources i The percentage of the number of objects provided to the total number of objects is obtained The calculation formula is as follows:
[0083]
[0084] Where N(s i , d j ) indicates the data source s i In the field j The number of objects provided in Indicates that in the field d j A collection of data sources that provide object description information.
[0085] Step 1-2: Introduce domain expertise in data sources Calculate. Among them, Indicates data source i In the field j Professionalism within The calculation formula is as follows:
[0086]
[0087] Where α is a hyperparameter for adjusting the weight of domain expertise. In the experiments on the dataset, α is set to 1.5.
[0088] Step 1-3: Import data sources i With object o k Domain similarity I(o k ,s i ) calculation. All data sources s1~s i Corresponding areas d1~d j Professionalism matrix As shown in the following formula:
[0089]
[0090] matrix The i-th row vector is defined as the data source s i The field vector According to the object k The domain construction object o k The field vector The calculation formula is as follows:
[0091]
[0092] Where: When object o k Belongs to field d j When otherwise Calculation data sources i With object o k Domain similarity I(o k ,s i )The formula is as follows:
[0093]
[0094] Step 1-4: Construction of fully connected domain explicit / implicit observation graph. j Data source collection In object collection The provided explicit / implicit observation relationship is modeled as a fully connected domain explicit / implicit observation graph Data Source For the picture The vertices in .
[0095] Step 1-5: Graph the explicit / implicit observations in the domain Introduce data sources i With data sources m Provides consensus of explicit / implicit observations of objects The calculation formula is as follows:
[0096]
[0097]
[0098] In the formula and Respectively represent data sources i and m In the field j a collection of objects providing observations, and Respectively represent data sources i and m In object o k The set of explicit observations provided, and Respectively represent data sources i and m In object o k The set of implicit observations provided, σ(v) and They represent the probability that the observed value v is true / false, respectively. and Indicates data source i With data sources m In object o k provides the similarity of explicit / implicit observations on and The calculation formula is as follows:
[0099]
[0100]
[0101] μ(si , o k )and Respectively represent data sources i In object o k The credibility score of the explicit observation provided on the attribute, data source s m In object o k The implicit observation value provided by the attribute is a credibility score, and the explicit / implicit observation value provided by the multi-truth data conflict data source is assigned a credibility score μ(s i , o k )and The calculation formula is as follows:
[0102]
[0103]
[0104] Where U(o k ) represents the object o k The complete set of observations.
[0105] Step 1-6: Graph the explicit / implicit observations in the domain Introduce data sources i For data sources m Long-tail compensation on explicit / implicit observations and The calculation formula is as follows:
[0106]
[0107]
[0108] In the formula I(o k ,s m ) indicates the data source s m With object o k The domain similarity of , σ(v) and They represent the probability that the observed value v is true / false, μ(s m , o k )and Respectively represent data sources m In object o k The credibility score of the explicit observation provided on the attribute, data source s m In object o k Provides a confidence score for the implicit observation on the attribute.
[0109] Step 1-7: Explicit / implicit observations in the domain map Inside, calculate the vertex s i Point to vertex s m Edge weight The calculation formula is as follows:
[0110]
[0111]
[0112] Where λ is the damping factor, ensuring the domain explicit / implicit observation value map has a stationary distribution, Indicates data source i With data sources m In the field j Provides the consensus of explicit / implicit observations, Indicates data source i For data sources m In the field j The long-tail compensation on explicit / implicit observations is provided.
[0113] Step 1-8: Introduce inter-vertex weight normalization calculation. and The calculation formula is as follows:
[0114]
[0115]
[0116] In the formula Represents the explicit / implicit observation value graph in the domain In the figure, vertex s i Point to vertex s m The weight of the edge.
[0117] Step 1-9: Introduce the reliability calculation of the data source domain. j In the fully connected domain, the explicit / implicit observation value graph Perform random walks to obtain the data source s in the explicit / implicit observation value graph i The stationary access probability p(s i )and Data Sources i In the field j Reliability within the field: Accuracy and false positive rate The calculation formula is as follows:
[0118]
[0119]
[0120] in and are the pre-set parameters. Represents the data source s in the explicit / implicit observation value diagram i The stationary access probability.
[0121] Step 2: Modeling the credibility of observations
[0122] The probability of the observation v being true / false is calculated based on the domain reliability of the data source and the relationship between the observations provided by the data source: The calculation formula is as follows:
[0123]
[0124]
[0125]
[0126]
[0127] In the formula Indicates data source i In the field j The accuracy / false positive rate within Indicates data source i In object o k The final accuracy / false positive rate in the field, and represents the set of data sources that provide explicit / implicit observations v, ″o k Represents object o k A collection of fields. Indicates that the object o is provided k The data source collection of observations.
[0128] Step 3: Truth Inference
[0129] when When , the observed value v is judged to be true, otherwise it is judged to be false, thus completing the true value inference.
[0130] In order to verify the accuracy and completeness of the truth value of this multi-truth value discovery method based on the reliability of the data source domain, taking the book dataset as an example, this multi-truth value discovery method based on the reliability of the data source domain is compared with the Majority Voting multi-truth value discovery method, the TruthFinder multi-truth value discovery method, the LTM multi-truth value discovery method, and the DART multi-truth value discovery method, as follows:
[0131] The book dataset used in the experiment by Xueling LIN et al. of the Hong Kong University of Science and Technology was adopted. The dataset contains 54,591 different registered booksellers, providing 2,338,559 pieces of information for 210,206 books, each of which is identified by ISBN-13. The attributes of books include "author", "publication year" and "category". The original data contains noise, and the "author" attribute is preprocessed to generate a unified format. The fields are divided according to the "category" and "publication year" attributes of the book. In the "category" attribute, each discrete value is a field. In the "publication year" attribute, the fields are divided according to the publication year time period containing the same number of objects. 100 books were randomly selected and the author information was manually annotated.
[0132] The indicators for evaluating the accuracy and completeness of the multi-truth discovery method include: recall (the average percentage of predicted actual truth values in the actual truth value set of all objects), precision (the average percentage of predicted actual truth values in the truth value set predicted by the algorithm) and harmonic mean F-measure (the harmonic mean of precision and recall). The experimental results are shown in Table 2 below.
[0133] Table 2 Comparison of different multi-truth discovery methods
[0134] Method Category Recall Precision F1 Majority Voting 0.712 0.87 0.783 TruthFinder 0.828 0.813 0.82 LTM 0.922 0.77 0.839 DART 0.973 0.575 0.723 Method of the present invention 0.83 0.864 0.847
[0135] As shown in the above table, the recall rate, precision rate and harmonic mean of this multi-truth value discovery method based on the reliability of the data source domain are at a relatively high level, and can be applied to many scenarios in the real world, such as big data fusion systems, construction of high-quality knowledge bases, crowd perception, privacy protection, crowdsourcing aggregation, and medical fields.
Claims
1. A multi-truth discovery method based on the reliability of data source domain, characterized in that: The specific steps include: Step 1: Data source domain reliability modeling; Step 1-1: Calculate the domain richness of the data source in For data sources i In the field j Provides data richness within the domain by calculating the domain d j Internal data sources i The percentage of the number of objects provided to the total number of objects is obtained The calculation formula is as follows: Where N(s i ,d j ) indicates the data source s i In the field j The number of objects provided in Indicates that in the field d j A collection of data sources that provide object description information; Step 1-2: Calculate the data source s i In the field j Professionalism within Step 1-3: Calculate the data source s i With object o k Domain similarity I(o k ,s i ), all data sources s1~s i Corresponding areas d1~d j Professional composition matrix The formula is as follows: According to the object k The domain construction object o k The field vector The formula is as follows: In the formula, when the object o k Belongs to field d j When otherwise Calculation data sources i With object o k Domain similarity I(o k ,s i )The formula is as follows: In the formula For data sources i The domain vector of The i-th row vector of ; Step 1-4: Construction of fully connected domain explicit / implicit observation graph, in domain d j Data source collection In object collection The provided explicit / implicit observation relationship is modeled as a fully connected domain explicit / implicit observation graph Data Source For the picture The vertices in Step 1-5: Graph the explicit / implicit observations in the domain In, calculate the data source s i With data sources m Provides consensus of explicit / implicit observations of objects Step 1-6: Graph the explicit / implicit observations in the domain In, calculate the data source s i For data sources m Long-tail compensation on explicit / implicit observations and The calculation formula is as follows: In the formula I(o k ,s m ) indicates the data source s m With object o k The domain similarity of , σ(v) and They represent the probability that the observed value v is true / false, μ(s m ,o k )and Respectively represent data sources m In object o k The credibility score of the explicit observation provided on the attribute, data source s m In object o k Provides credibility scores for implicit observations on attributes; Step 1-7: Explicit / implicit observations in the domain map Inside, calculate the vertex s i Point to vertex s m Edge weight Step 1-8: Normalization calculation of inter-vertex weights, calculation of inter-vertex normalized weights and Step 1-9: Calculate the reliability of the data source domain, in domain d j Explicit / implicit observation graph of fully connected domains Perform random walks to obtain the data source s in the explicit / implicit observation graph respectively i The stationary access probability p(s i )and Calculation data sources i In the field j The accuracy within and false positive rate Get data sources i In the field j Reliability within the domain; Accuracy and false positive rate The calculation formula is as follows: In the formula and are the pre-set parameters. Represents the data source s in the explicit / implicit observation value diagram i The stationary access probability of Step 2: Observation credibility modeling; Calculate the probability of the observation v being true / false based on the domain reliability of the data source and the relationship between the observations provided by the data source The calculation formula is as follows: In the formula Indicates data source i In the field j The accuracy / false positive rate within Indicates data source i In object o k The final accuracy / false positive rate in the field, and represents the set of data sources that provide explicit / implicit observations v, Represents object o k A collection of fields. Indicates that the object o is provided k The data source collection of observations; Step 3: True value inference; when When , the observed value v is judged to be true, otherwise it is judged to be false, completing the true value inference.
2. The method for discovering multiple truth values based on the reliability of data source domain according to claim 1, characterized in that: Calculate the data source s in step 1-2 i In the field j Professionalism within The calculation formula is as follows: Where α is a hyperparameter for adjusting the weight of domain expertise. In the experiments on the dataset, α is set to 1.
5.
3. The method for discovering multiple truth values based on the reliability of data source domain according to claim 1, characterized in that: In steps 1-5, in the domain explicit / implicit observation value diagram In, data sources i With data sources m Provides consensus of explicit / implicit observations of objects The calculation formula is as follows: In the formula and Respectively represent data sources i and m In the field j a collection of objects providing observations, and Respectively represent data sources i and m In object o k The set of explicit observations provided, and Respectively represent data sources i and m In object o k The set of implicit observations provided, σ(v) and They represent the probability that the observed value v is true / false, respectively. and Indicates data source i With data sources m In object o k provides the similarity of explicit / implicit observations, and The calculation formula is as follows: μ(s i ,o k )and Respectively represent data sources i In object o k The credibility score of the explicit observation provided on the attribute, data source s m In object o k The implicit observation value provided on the attribute has a credibility score, and the explicit / implicit observation value provided by the multi-truth data conflict data source is assigned a credibility score μ(s i ,o k )and The calculation formula is as follows: Where U(o k ) represents the object o k The complete set of observations.
4. The method for discovering multiple truth values based on the reliability of data source domain according to claim 1, characterized in that: Calculate the vertex s in steps 1-7 i Point to vertex s m Edge weight The calculation formula is as follows: Where λ is the damping factor, ensuring the domain explicit / implicit observation value map has a stationary distribution, Indicates data source i With data sources m In the field j Provides the consensus of explicit / implicit observations, Indicates data source i For data sources m In the field j Provides long-tail compensation on explicit / implicit observations.
5. The method for discovering multiple truth values based on the reliability of data source domain according to claim 1, characterized in that: Normalized weights between vertices in steps 1-8 and The calculation formula is as follows: In the formula Represents the explicit / implicit observation value graph in the domain Midpoints i Point to vertex s m The weight of the edge.
