Information processing program, information processing method and information processing device

Abstract

To provide an information processing program, an information processing method and an information processing device for facilitating the confirmation of credibility of received information.SOLUTION: An information processing program has a learned credibility determination tendency model 114 for outputting the credibility of data with property information showing the certainty of the data and having confidence information showing the confidence of the property information added thereto as input, and causes a computer to execute processing for acquiring first property information showing the certainty of prescribed data received by a recipient, generate first confidence information showing the confidence of the first property information on the basis of a trust list 112 registering a generation subject trusted by the recipient among generation subjects of the first property information, input the first property information added with the first confidence information to the credibility determination tendency model 114 and present a result obtained by determining the credibility of the prescribed data.SELECTED DRAWING: Figure 11

Description

[Technical field]

[0001] The present invention relates to an information processing program, an information processing method, and an information processing device. [Background technology]

[0002] The Internet is overflowing with various information, such as articles on news sites and information provided through SNS (Social Networking Services). This information on the Internet can contain false information, which is information that contains incorrect content. In recent years, such false information has become a problem because it can cause chaos in the economy and society.

[0003] In response to this, a technology has been proposed to universally verify the authenticity of data on the Internet as a countermeasure against false information. In this technology, property information that represents an evaluation regarding verification and evaluation by a third party other than the sender and receiver of the data is added to data on the Internet. Then, the property information and information of the provider, who is the third party who provided the property information, are graphed and disclosed to the receiver. This allows the receiver to determine the authenticity of the data to which the property information is added, thereby increasing the reliability of the data. In the graph, for example, information of the sender of the data and the provider of the property information are each represented as a node corresponding to an element related to the data.

[0004] Furthermore, the following techniques can be considered as methods of using such graph-based technology. For example, a machine learning model is trained using training data including data whose veracity is clear and property information. Then, unknown data and property information for that data can be input to the trained machine learning model, and the model can be made to output the veracity of the data.

[0005] As a technology related to the credibility of information, a technology has been proposed in which a machine learning model is trained using the rating reliability and rating trends of multiple raters to determine the credibility of news rated by the raters. Another technology has been proposed in which a machine learning model is trained on the posting history of a poster and the trained machine learning model is used to determine fraudulent posting of product reviews. Another technology has been proposed in which a report is received, the category of the report is identified, and the credibility of the report is evaluated using a machine learning model corresponding to the identified category. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Special Publication No. 2020-508518 [Patent Document 2] JP 2022-147013 A [Patent Document 3] US Patent Application Publication No. 2021 / 0289166 Summary of the Invention [Problem to be solved by the invention]

[0007] However, in a technology in which unknown data and property information are input into a machine learning model to obtain a judgment result of authenticity, the judgment is made without clear how trustworthy the provider of each piece of property information is. Furthermore, how trustworthy a provider of property information is varies depending on the recipient and the object of judgment.

[0008] For example, let us consider a case where a receiver wants to confirm the credibility of a description that a specialized economics textbook is "easy to understand." In this case, it is conceivable that an economics expert would evaluate the specialized textbook as easy to understand, while a person who has just started studying economics would evaluate it as difficult to understand. In this situation, if the receiver of the description has some knowledge of economics, the receiver will trust the opinion of the expert more than a person who has just started studying economics, and will perceive the description as true. On the other hand, if the receiver of the description is a beginner in economics, the receiver will trust the opinion of a person who has just started studying economics more than an expert, and will perceive the description as false. In this way, the reliability of the provider of property information changes depending on the attributes of the receiver.

[0009] Therefore, in a technology in which unknown data and property information are input into a machine learning model to obtain a judgment result of authenticity, the reliability of the provider of the property information is not taken into consideration, and it is difficult to obtain an appropriate judgment result of authenticity for the receiver. Therefore, it is difficult to make an appropriate judgment of authenticity for the received information.

[0010] In addition, a technology that evaluates credibility by having a machine learning model learn using the evaluation reliability and evaluation trends of multiple evaluators is based on information from general evaluators, and it is difficult for a receiver who has obtained the data to obtain the credibility of the obtained data. In addition, a technology that determines whether a product review is fraudulent by having a machine learning model learn the poster's posting history, and a technology that evaluates credibility using a machine learning model according to the report category do not take into account the credibility of the obtained data for the receiver who has obtained the data. Therefore, it is difficult to appropriately determine the credibility of received information using either technology.

[0011] The disclosed technology has been made in consideration of the above, and aims to provide an information processing program, an information processing method, and an information processing device that enable appropriate credibility determination of received information. [Means for solving the problem]

[0012] In one aspect of the information processing program, information processing method, and information processing device disclosed in the present application, a computer is caused to execute a process of acquiring first property information indicating the reliability of specified data received by a receiver, generating first reliability information indicating the reliability of the first property information based on a reliability list in which generators of the first property information that the receiver trusts are registered, inputting the first property information with the first reliability information assigned to it into the first machine learning model, and determining the reliability of the specified data and presenting the result. Effect of the Invention

[0013] In one aspect, the present invention can enable appropriate determination of the authenticity of received information. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a system configuration diagram of an information processing system according to an embodiment. [Diagram 2] FIG. 2 is a block diagram of an information processing device according to an embodiment. [Diagram 3] FIG. 3 is a diagram illustrating an example of the graph information. [Figure 4] FIG. 4 is a diagram illustrating an example of the trust list. [Diagram 5] FIG. 5 is a diagram showing an example of the labeling of the reliability using the reliability list. [Figure 6] FIG. 6 is a first diagram showing an example of reliability labeling using a reliability determination model. [Figure 7] FIG. 7 is a second diagram showing an example of reliability labeling using the reliability judgment model. [Figure 8] FIG. 8 is a diagram showing an example of labeling the reliability of edges leading to judgment target information. [Figure 9] FIG. 9 is a diagram showing an example of credibility determination using the credibility determination tendency model. [Figure 10] FIG. 10 is a diagram illustrating an example of parameter update of the credibility determination tendency model. [Figure 11] FIG. 11 is a diagram illustrating an overall flow of the credibility determination process according to the embodiment. [Figure 12] FIG. 12 is a flowchart of the credibility determination process by the information processing device according to the embodiment. [Figure 13] FIG. 13 is a hardware configuration diagram of an information processing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an information processing program, an information processing method, and an information processing device disclosed in the present application will be described in detail with reference to the accompanying drawings. Note that the information processing program, the information processing method, and the information processing device disclosed in the present application are not limited to the following embodiments. EXAMPLES

[0016] Fig. 1 is a system configuration diagram of an information processing system according to an embodiment. Fig. 2 is a block diagram of an information processing device according to an embodiment. In Fig. 2, the Internet 4 is omitted in order to make the flow of communication easier to understand, but in reality, each device communicates via the Internet 4 as shown in Fig. 1.

[0017] The information processing system 100 includes an information processing device 1, a recipient terminal device 2, and a server group 3. The information processing device 1, the recipient terminal device 2, and the server group 3 are each connected to the Internet 4 and are capable of transmitting and receiving data to and from each other.

[0018] The server group 3 includes a reliability determination model management server 31, a credibility determination trend model management server 32, a trust list management server 33, a graph information management server 34, and an SNS server 35. The SNS server 35 is a device that provides information to be determined for authenticity. The recipient determines the credibility of the information provided by the SNS server 35. To make this determination, the recipient uses a credibility determination process by the information processing device 1, which will be described later. The server that provides the information to be determined for authenticity may be a server other than the SNS server 35.

[0019] The graph information management server 34 receives an input of information for which the receiver judges the credibility from the information processing device 1. Hereinafter, the information for which the receiver judges the credibility is referred to as "judgment target information". Then, the graph information management server 34 acquires various information on the Internet 4, including information provided by the SNS server 35, and creates graph information 111 for the judgment target information, which has the judgment target information as its root (root). Herein, a hierarchical graph is registered in the graph information 111 by repeatedly branching from one starting point, and the starting point of the hierarchy in the graph is called the root. That is, the graph information 111 has a hierarchical structure or a tree structure with the root as its starting point. Then, the graph information management server 34 transmits the created graph information 111 for the judgment target information to the information processing device 1.

[0020] The reliability determination model management server 31, the credibility determination tendency model management server 32, and the trust list management server 33 are servers that manage information owned by the recipient. In Fig. 1, the reliability determination model management server 31, the credibility determination tendency model management server 32, and the trust list management server 33 are described as being connected to the Internet 4, but may be arranged in the local environment of the recipient, for example. The reliability determination model management server 31, the credibility determination tendency model management server 32, and the trust list management server 33 provide various information related to the recipient to the information processing device 1 via the Internet 4.

[0021] The receiver terminal device 2 is a terminal device operated by a receiver who judges the credibility of information provided from the SNS server 35. The receiver terminal device 2 notifies the information processing device 1 of the designation of the judgment target information. In addition, upon receiving the designation from the receiver, the receiver terminal device 2 notifies the information processing device 1 of the designation of a trust list 112, a trust degree determination model 113, and a credibility determination tendency model 114 corresponding to the receiver. The details of the trust list 112, the trust degree determination model 113, and the credibility determination tendency model 114 will be described later.

[0022] The receiver terminal device 2 receives the judgment result of the credibility of the judgment target information and the graph information 111 that is used to judge the credibility from the information processing device 1. Then, the receiver terminal device 2 provides the information to the receiver by displaying it on a monitor or the like.

[0023] Thereafter, if there is information to be fed back from the receiver to the information processing device 1 in response to the judgment result of the credibility of the information processing device 1, the receiver terminal device 2 transmits the designated feedback information 20 to the information processing device 1. Details of the feedback information 20 will be described later.

[0024] The information processing device 1 executes a credibility judgment process for the judgment target information according to the recipient and the judgment target information, using information acquired from the reliability judgment model management server 31, the credibility judgment trend model management server 32, the trust list management server 33, and the graph information management server 34. The information processing device 1 also generates unique graph information specific to the recipient to which information serving as a basis for judging the credibility of the judgment target information is added. Then, the information processing device 1 transmits the judgment result of the credibility of the judgment target information and the unique graph information to the recipient terminal device 2.

[0025] The credibility determination process by the information processing device 1 will be described in detail below with reference to Fig. 2. As shown in Fig. 2, the information processing device 1 includes a data storage unit 11, a unique graph information generation unit 12, a credibility determination unit 13, an output unit 14, and a machine learning model update unit 15.

[0026] The data storage unit 11 receives notification of designation of judgment target information, a trust list 112 corresponding to the recipient, a reliability judgment model 113, and a credibility judgment tendency model 114 from the recipient terminal device 2. For example, the judgment target information is input to the data storage unit 11 by a drag operation using a mouse or the like to a part of the text of the SNS post displayed on the monitor of the recipient terminal device 2 that is to be judged for credibility. Information on which of predetermined topics the judgment target information belongs to is added to the judgment target information. Information on property information for the judgment target information may also be added to the judgment target information.

[0027] Then, the data storage unit 11 instructs the graph information management server 34 of the server group 3 to create graph information 111 for the judgment target information. Thereafter, the data storage unit 11 receives, from the graph information management server 34, the graph information 111 for the judgment target information in which a graph having a hierarchical structure with the judgment target information specified by the receiver as the root is registered. Then, the data storage unit 11 stores the received graph information 111 for the judgment target information.

[0028] 3 is a diagram showing an example of graph information. The graph information 111 is information having a graph structure including, for example, judgment target information 120, a plurality of pieces of property information 121, and a plurality of entities 122 indicating the respective generating subjects of the property information 121. The property information 121 is data in which properties, which are information on attributes and functions of data or an individual, a corporation, a device, etc., are recorded in a format that cannot be tampered with by a digital signature. The property information 121 may include information indicating the generating subject.

[0029] As shown in FIG. 3, the graph information 111 is registered with information represented by a hierarchical structure with the judgment target information 120 received by the receiver as the root. In the graph information 111, for example, three pieces of property information 121 are associated with the judgment target information 120 received by the receiver. Then, an entity 122 is associated with each of the three pieces of property information. For example, an entity 122 indicating that the person is a river department staff member is associated with the property information 121 of "River A: no disaster warning". Furthermore, when property information 121 exists for each entity 122, each is associated with the property information. Furthermore, an entity 122 is associated with the property information 121 for each entity 122. For example, the property information 121 of "staff ID: river department" is associated with the entity 122 indicating the river staff member. Then, the entity 122 indicating "City B" is associated with the entity 122 indicating the property information 121 as the generating subject. In this way, the graph information 111 has a hierarchical structure in which the judgment target information 120 is the root, which is the starting point of the hierarchical structure, and the property information 121 and the entities 122 that are the generating subjects of each piece of property information 121 are linked together.

[0030] By looking at this graph information 111, the user can confirm what kind of evaluation was made by which generating entity for the judgment target information 120. However, the graph information 111 in this state represents the same generic information for everyone, and although the user can simply confirm the contents of the property information 121 related to the judgment target information 120 and its generating entity, it is difficult for the user to grasp how much each generating entity can be trusted.

[0031] 2, the description will be continued. The data holding unit 11 receives the trust list 112 specified by the recipient from the trust list management server 33 of the server group 3. Then, the data holding unit 11 stores the received trust list 112.

[0032] FIG. 4 is a diagram showing an example of a trust list. As shown in FIG. 4, a trust list 112 is created for each receiver and for each topic to be judged. The trust list 112 is a list of generators that the receiver considers to be trustworthy for property information in a specific topic. The trust list 112 describes the topic of the information to be judged and a unique identifier of the generator that the receiver considers to be trustworthy. For example, a DID (Decentralized Identifier) ​​can be used as the unique identifier. In this embodiment, information indicating who the generator is is added to the property information using the same type of identifier as that handled in the trust list 112.

[0033] For example, the trust list 112 in Fig. 4 is a list for the topic "disasters." In the trust list 112 in Fig. 4, the recipient is designated as a generator trusted by City B, the local government in which the recipient lives, and the Global Positioning System (GPS), which issues location information for water level sensors, etc. For example, syntax 130 is an example of a document in the trust list 112 using DID.

[0034] 2, the description will be continued. The data holding unit 11 receives the reliability judgment model 113 specified by the recipient from the reliability judgment model management server 31 of the server group 3. Then, the data holding unit 11 stores the received reliability judgment model 113.

[0035] The reliability determination model 113 is a machine learning model that determines the reliability of property information registered in the graph information 111 by using a part of the graph information 111 including a path that can be traced from the determination target information, which is the root, to the property information. Here, the part of the graph information 111 that can be traced from the determination target information, the reliability of which is to be determined, is called "subgraph information." For example, in FIG. 3, information that includes the path from the determination target information 120 to the entity 122 indicating the GPS via the entity 122 indicating the water level sensor is an example of subgraph information.

[0036] That is, the reliability determination model 113 is used when determining the reliability of property information from a specific generating entity not specified in the reliability list 112 in the graph information 111 based on the relationship with other generating entities. Specifically, the reliability determination model 113 determines the reliability of property information obtained from a specific generating entity by learning what kind of property information has been issued to the specific generating entity by a generating entity determined by the receiver to have high reliability. For example, the reliability determination model 113 can use the receiver's own past judgments about the reliability of property information related to the same type of judgment target information, such as the same topic, for machine learning.

[0037] The reliability determination model 113 is an example of a "second machine learning model." Furthermore, property information included in the learning data used in the machine learning of the reliability determination model 113 is an example of "second property information." In other words, the reliability determination model 113 is a machine learning model that receives specific property information and information about a generator of the specific property information as input and outputs reliability information of the specific property information, and machine learning is performed based on past reliability judgment results for the second property information of the recipient.

[0038] 2, the description will be continued. The data holding unit 11 receives the credibility determination tendency model 114 specified by the receiver from the credibility determination tendency model management server 32 of the server group 3. Then, the data holding unit 11 stores the received credibility determination tendency model 114.

[0039] The credibility judgment tendency model 114 is a machine learning model that receives input of unique graph information, which is graph information unique to the receiver for the judgment target information, and outputs a judgment result of the credibility of the judgment target information. The judgment result of the credibility is expressed, for example, as information such as true, false, or unknown, or a numerical value indicating the degree of credibility. The credibility judgment tendency model 114 performs machine learning using past judgments of credibility made by the receiver himself / herself in the same topic as the judgment target information. The credibility judgment tendency model 114 can use, for example, a Graph Neural Network (GNN) or the like that can use a graph as an input. The credibility judgment tendency model 114 is an example of a "first machine learning model."

[0040] The unique graph information generation unit 12 acquires the graph information 111, the trust list 112, and the reliability determination model 113 from the data storage unit 11. Then, the unique graph information generation unit 12 uses the trust list 112 and the reliability determination model 113 to generate unique graph information in which reliability information is added to each piece of property information in the graph information 111. Then, the unique graph information generation unit 12 outputs the generated unique graph information to the credibility determination unit 13 and the output unit 14. Details of the creation of unique graph information by the unique graph information generation unit 12 will be described below.

[0041] The unique graph information generating unit 12 assigns a label indicating that the property information from the generating subject listed in the trust list 112 in the graph information 111 is “highly reliable.” Fig. 5 is a diagram showing an example of the labeling of the reliability using the trust list.

[0042] For example, consider a case where City B, which is the local government where the receiver lives, is registered in the trust list 112, but a GPS that issues position information of a water level sensor or the like as property information is not registered. The unique graph information generation unit 12 assigns "H" as a label indicating high reliability to the edge 141 extending from the entity 122a representing City B to the property information 121a in the graph information 111 of Fig. 5 (step S1). In this case, since the GPS is not registered in the trust list 112, the unique graph information generation unit 12 does not label the edge extending from the GPS to the property information in the graph information 111 of Fig. 5.

[0043] Here, the judgment target information is an example of "predetermined data". Moreover, the property information for the judgment target information in the graph information 111 is an example of "first property information". Moreover, a label indicating the reliability attached to an edge extending from a generating subject registered in the trust list 112 to the property information is an example of "first reliability information". That is, the unique graph information generating unit 12 generates first reliability information indicating the reliability of the first property information based on the trust list 112 in which a generating subject trusted by the receiver among the generating subjects of the first property information is registered. Moreover, the unique graph information generating unit 12 acquires the first property information by acquiring graph information 111 indicating the relationship between the predetermined data, the first property information, and the generating subject.

[0044] In the following description, in the graph information 111, the label "H" indicates high reliability, and the label "L" indicates low reliability. In this embodiment, the unique graph information generating unit 12 performs labeling with information indicating whether the reliability is high or low, but a numerical value indicating the reliability may be used as the label.

[0045] Next, for property information by a generator not listed in the trust list 112, the unique graph information generating unit 12 inputs subgraph information rooted at the determination target information for which the trust is to be determined into the trust determination model 113, and assigns a label indicating the trust determination result to be output. Fig. 6 is a first diagram showing an example of trust labeling using the trust determination model. Fig. 7 is a second diagram showing an example of trust labeling using the trust determination model.

[0046] The unique graph information generation unit 12 extracts subgraph information 142 including a trace from the judgment target information 120 to the GPS entity 122b not registered in the trust list 112 via the water level sensor entity 122c from the graph information 111 shown in Fig. 6. Next, the unique graph information generation unit 12 inputs the subgraph information 142 to the reliability determination model 113 (step S2). Then, the unique graph information generation unit 12 obtains a reliability determination result output from the reliability determination model 113. Here, the unique graph information generation unit 12 obtains an output 143 indicating that both property information provided by the GPS indicated by the entity 122b and the water level sensor indicated by the entity 122c of the subgraph information 142 have high reliability (step S3).

[0047] 7, the unique graph information generation unit 12 assigns a label "H" indicating high reliability to the edge 144 extending from the entity 122b to the property information 121b. The unique graph information generation unit 12 also assigns a label "H" indicating high reliability to the edge 145 extending from the entity 122c to the property information 121c (step S4).

[0048] Here, for example, the GPS corresponds to a "generation subject not registered in the trust list 112." The property information 121b provided by the GPS and the property information 121c provided by the water level sensor to which the property information 112b is added are an example of "third property information and information on the generation subject of the third property information." The labels added to the edges 144 and 145 are an example of "first reliability information." That is, the unique graph information generation unit 12 inputs the third property information and information on the generation subject of the third property information into the reliability determination model 113, which is the second machine learning model, to generate the first reliability information.

[0049] Next, the unique graph information generating unit 12 uses the reliability judgment model 113 to judge the reliability labels for the property information indicating each generating subject, starting from the edge farthest from the judgment target information in the graph information 111, and assigns a label indicating reliability to each edge. Here, by assigning labels in order starting from the edge farthest from the judgment target information, the unique graph information generating unit 12 can perform reliability judgment using information on the labels assigned to the edges below the label assignment target. The unique graph information generating unit 12 generates unique label information by assigning reliability labels to all edges in the graph information 111. FIG. 8 is a diagram showing an example of reliability labeling of edges leading to the judgment target information.

[0050] For example, as shown in FIG. 8, the unique graph information generating unit 12 confirms that the label representing the reliability of the edge 141 is "H". Next, the unique graph information generating unit 12 inputs the label information of the edge 141 and the subgraph information including the entities 122d and 122a to the reliability determination model 113. Then, the unique graph information generating unit 12 acquires the judgment result of the reliability of the property information 121d provided from the entity 122d representing the River Section staff as an output for the input. Here, the unique graph information generating unit 12 acquires "H" as a label representing the reliability of the property information 121d and assigns it to the edge 146. In addition, the unique graph information generating unit 12 inputs the subgraph information including the property information 121e to the reliability determination model 113 to acquire the judgment result of the reliability of the property information 121e provided from the entity 122e representing a person pretending to be a River Section staff. In this case, since there is no edge extending to the entity 122e representing the person impersonating a river department employee, the unique graph information generating unit 12 does not input label information for the non-existent edge. Here, the unique graph information generating unit 12 obtains "L" as a label indicating the reliability of the property information 121e provided by the entity 122e, and assigns it to the edge 147 (step S5).

[0051] Returning to FIG. 2, the explanation will be continued. The credibility determination unit 13 acquires the credibility determination tendency model 114 from the data storage unit 11. In addition, the credibility determination unit 13 receives an input of unique graph information of the recipient from the unique graph information generation unit 12. Next, the credibility determination unit 13 inputs the unique graph information to the credibility determination tendency model 114, and acquires a credibility determination result output from the credibility determination tendency model 114. Thereafter, the credibility determination unit 13 outputs the acquired credibility determination result to the output unit 14. In addition, the credibility determination unit 13 outputs the credibility determination result for the diagnostic target information to the machine learning model update unit 15.

[0052] 9 is a diagram showing an example of credibility determination using a credibility determination tendency model. The credibility determination unit 13 inputs the individual graph information 150 acquired from the unique graph information generation unit 12 to the credibility determination tendency model 114 (step S6). Then, the credibility determination unit 13 acquires the judgment result 148 of the credibility of the judgment target information input from the credibility determination tendency model 114 (step S7). For example, as shown in FIG. 9, the credibility determination unit 13 acquires information indicating any one of "true", "unknown" or "false" as the judgment result 148 of the credibility of the judgment target information.

[0053] Returning to Fig. 2, the explanation will be continued. The output unit 14 receives an input of the recipient's unique graph information from the unique graph information generation unit 12. The output unit 14 also receives an input of the judgment result of the credibility of the judgment target information from the credibility judgment unit 13. Then, the output unit 14 transmits the recipient's unique graph information and the judgment result of the credibility of the judgment target information to the recipient terminal device 2, and provides them to the recipient.

[0054] The receiver uses the receiver terminal device 2 to check the judgment result of the credibility of the provided unique graph information and the judgment target information, and judges the credibility of the judgment target. After that, the receiver uses the receiver terminal device 2 or the like to generate credibility judgment result information 21 indicating the judgment result of the credibility finally made for the judgment target. The receiver also uses the receiver terminal device 2 or the like to generate virtual evaluation information 22, which is a proposal of property information that is not currently added to the acquired unique graph information but that the receiver would judge to have high credibility of the judgment target if it were added. In addition, if there is a label to correct the value of the credibility label attached to the acquired unique graph information, the receiver generates reliability label correction information 23 indicating the label and the correction value. Then, the receiver terminal device 2 transmits the credibility judgment result information 21, the virtual evaluation information 22, and the reliability label correction information 23 to the information processing device 1 as feedback information 20.

[0055] In addition, the output unit 14 receives a request to transmit the updated reliability determination model 113 and the credibility determination tendency model 114 from the machine learning model update unit 15. Then, the output unit 14 transmits the updated reliability determination model 113 and the credibility determination tendency model 114 to the receiver terminal device 2.

[0056] The machine learning model update unit 15 receives feedback information 20 including credibility judgment result information 21, virtual evaluation information 22, and reliability label correction information 23 from the receiver terminal device 2. The machine learning model update unit 15 also receives an input of the judgment result of the credibility of the diagnostic target information from the credibility judgment unit 13. Then, the machine learning model update unit 15 uses the correction information of the label specified in the reliability label correction information 23 to update the parameters of the reliability judgment model 113 so as to reduce the error between the output of the reliability judgment model 113 and the judgment of the receiver.

[0057] Furthermore, the machine learning model update unit 15 compares the credibility judgment result registered in the credibility judgment result information 21 with the credibility of the judgment target information judged by itself, and further updates the parameters of the credibility judgment tendency model 114 by using the virtual evaluation information 22. In this case, the machine learning model update unit 15 updates the parameters of the credibility judgment tendency model 114 so as to reduce the error between the output of the credibility judgment tendency model 114 and the judgment of the receiver.

[0058] For example, a case will be described in which the reliability determination model 113 and the credibility determination tendency model 114 are modeled as a neural network such as a GNN. In this case, the machine learning model update unit 15 can update the parameters using a method based on the error between the output of the neural network and the receiver's judgment, which is the correct label, such as Stochastic Gradient Descent (SGD) or Adam.

[0059] 10 is a diagram showing an example of parameter update of the credibility determination tendency model. The machine learning model update unit 15 acquires a judgment result 149 in which the credibility output from the credibility determination tendency model 114 is determined to be "true". In addition, the machine learning model update unit 15 acquires credibility judgment result information 21 in which the credibility is determined to be "false" by the receiver P. Next, the machine learning model update unit 15 compares and judges the judgment result 149 and the credibility judgment result information 21 (step S8). Then, the machine learning model update unit 15 updates the parameters of the credibility determination tendency model 114 using the comparison result and the virtual evaluation information 22 (step S9).

[0060] Here, in this embodiment, the reliability of the credibility judgment by the information processing device 1 can be improved by updating the parameters of the credibility judgment tendency model 114 using the virtual evaluation information 22 fed back from the receiver. However, the virtual evaluation information 22 is an example of data for increasing the reliability of the credibility judgment fed back from the receiver, and other data may be used as long as it can contribute to improving the reliability of the credibility judgment. In addition, even if special data for increasing the reliability of the credibility judgment such as the virtual evaluation information 22 is not used, the machine learning model update unit 15 can update the parameters of the credibility judgment tendency model 114 based on the credibility judgment result information 21. This virtual evaluation information 22 is an example of "virtual property information".

[0061] Thereafter, the machine learning model update unit 15 outputs the updated reliability judgment model 113 and the credibility judgment tendency model 114 to the output unit 14, and requests transmission to the receiver terminal device 2. The receiver uses the receiver terminal device 2 to transmit the received updated reliability judgment model 113 to the reliability judgment model management server 31, thereby updating the receiver's own reliability judgment model 113 held by the reliability judgment model management server 31. Similarly, the receiver uses the receiver terminal device 2 to transmit the received updated reliability judgment tendency model 114 to the credibility judgment tendency model management server 32, thereby updating the receiver's own credibility judgment tendency model 114 held by the credibility judgment tendency model management server 32.

[0062] Here, in this embodiment, the receiver is made to update the reliability judgment model 113 and the credibility judgment trend model 114 held by the reliability judgment model management server 31 and the credibility judgment trend model management server 32, but this update may also be performed by the machine learning model update unit 15.

[0063] 11 is a diagram showing an overall flow of the credibility determination process according to the embodiment. Next, the overall picture of the credibility determination process by the information processing device 1 will be summarized again with reference to FIG.

[0064] In order to judge the credibility of the judgment target information obtained from the SNS server 35, the receiver inputs the judgment target information, the trust list 112, the reliability judgment model 113, and the credibility judgment tendency model 114 to the information processing device 1 using the receiver terminal device 2. The data holding unit 11 receives the trust list 112 specified by the receiver from the trust list management server 33 and stores it. The data holding unit 11 also acquires the reliability judgment model 113 specified by the receiver from the reliability judgment model management server 31 and stores it. The data holding unit 11 also acquires the credibility judgment tendency model 114 specified by the receiver from the credibility judgment tendency model management server 32 and stores it. The data holding unit 11 also requests the graph information management server 34 to instruct the graph information management server 34 to create the graph information 111 for the input judgment target information. After that, the data holding unit 11 receives and stores the graph information 111 for the judgment target information created by the graph information management server 34. The unique graph information generating unit 12 generates unique graph information 151 by using the graph information 111, the reliability list 112, and the reliability determination model 113 stored in the data holding unit 11 (step S11).

[0065] The credibility determination unit 13 inputs the unique graph information 151 generated by the unique graph information generation unit 12 to the credibility determination tendency model 114 (step S12). Then, the credibility determination unit 13 acquires the credibility determination result 152 output from the credibility determination tendency model 114.

[0066] The output unit 14 transmits the credibility determination result 152 acquired from the credibility determination unit 13 to the recipient terminal device 2 (step S13). The output unit 14 also transmits the unique graph information 151 generated by the unique graph information generation unit 12 to the recipient terminal device 2 (step S14). The recipient terminal device 2 provides result information 153 including the unique graph information 151 and the credibility determination result 152 to the recipient P by, for example, displaying it on a monitor.

[0067] The receiver P checks the result information 153 and judges the credibility of the judgment target information. After that, the receiver P transmits the feedback information 20 including the credibility judgment result information 21, the virtual evaluation information 22, and the reliability label correction information 23 to the machine learning model update unit 15 of the information processing device 1 (step S15).

[0068] The machine learning model update unit 15 updates the parameters of the credibility determination tendency model 114 using the credibility determination result information 21 and the virtual evaluation information 22 included in the feedback information 20 (step S16). In addition, the machine learning model update unit 15 updates the parameters of the credibility determination model 113 using the credibility label correction information 23 included in the feedback information 20 (step S17). At the next time, the information processing device 1 executes the credibility determination process using the credibility determination model 113 and the credibility determination tendency model 114 whose parameters have been updated.

[0069] 12 is a flowchart of the credibility determination process by the information processing device according to the embodiment. Next, a flow of the credibility determination process by the information processing device 1 according to the embodiment will be described with reference to FIG.

[0070] The data holding unit 11 receives inputs of designation of the judgment target information, the reliability list 112, the reliability judgment model 113, and the credibility judgment tendency model 114 from the receiver terminal device 2 (step S101).

[0071] Next, the data storage unit 11 acquires the specified reliability judgment model 113, credibility judgment trend model 114, and reliability list 112 from the reliability judgment model management server 31, the credibility judgment trend model management server 32, and the reliability list management server 33 (step S102).

[0072] The data storage unit 11 also issues an instruction to create graph information 111 for the input judgment target information to the graph information management server 34. After that, the data storage unit 11 acquires the graph information 111 for the judgment target information from the graph information management server 34 (step S103).

[0073] The unique graph information generating unit 12 generates unique graph information 151 using the graph information 111, the trust list 112, and the trust level determination model 113 stored in the data holding unit 11 (step S104).

[0074] The credibility determination unit 13 inputs the unique graph information 151 generated by the unique graph information generation unit 12 to the credibility determination tendency model 114 and obtains the output, thereby determining the credibility of the diagnostic object information (step S105).

[0075] The output unit 14 outputs the unique graph information generated by the unique graph information generation unit 12 and the credibility judgment result acquired from the credibility judgment unit 13 to the receiver terminal device 2 (step S106). The receiver P checks the result information 153 and judges the credibility of the judgment target information.

[0076] After that, the machine learning model update unit 15 determines whether or not there is feedback from the recipient (step S107). If there is no feedback (step S107: No), the machine learning model update unit 15 ends the process.

[0077] On the other hand, if there is feedback (step S107: yes), the machine learning model update unit 15 then acquires feedback information 20 including credibility judgment result information 21, virtual evaluation information 22, and reliability label correction information 23 from the receiver terminal device 2 (step S108).

[0078] Furthermore, the machine learning model update unit 15 updates the parameters of the reliability determination model 113 using the reliability label correction information 23 included in the feedback information 20 (step S109).

[0079] Next, the machine learning model update unit 15 updates the parameters of the credibility judgment tendency model 114 using the credibility judgment result information 21 and the virtual evaluation information 22 included in the feedback information 20 (step S110).

[0080] After that, the output unit 14 transmits the reliability determination model 113 and the credibility determination tendency model 114 updated by the machine learning model update unit 15 to the receiver terminal device 2 (step S111).

[0081] As described above, the information processing device according to this embodiment uses a trust list and a reliability judgment model to judge the reliability of property information by each generating entity shown in the graph information of the judgment target information, and generates unique graph information by adding a label indicating the reliability to the graph information.The information processing device then judges the reliability of the judgment target information using the generated unique graph information in a credibility judgment tendency model.The receiver can judge the credibility of the diagnosis target information based on the judgment result of the credibility of the diagnosis target information and the unique graph information.Furthermore, the information processing device acquires feedback information including information on the result of the receiver's judgment of credibility, and updates the reliability judgment model and the credibility judgment tendency model.

[0082] Since the reliability judgment model performs machine learning using the reliability judgment result fed back from the receiver, the reliability of the property information from each generating entity can be judged according to the judgment tendency of the receiver. Then, by labeling the graph information with the reliability using the information of the generating entities trusted by the receiver listed in the trust list and the reliability judgment model, it is possible to generate unique graph information that indicates the reliability according to the judgment tendency of the receiver. Furthermore, by judging the reliability of the diagnostic target information using the unique graph information, it is possible to judge the reliability of the judgment target by the unique graph information reflecting the reliability of each generating entity from the receiver's perspective, and to obtain a reliability judgment result according to the judgment tendency of the receiver. As a result, when using the graph information to support the judgment of reliability for data whose reliability received by the receiver is unknown, it is possible to support the judgment of the reliability of the data in consideration of the reliability of the generating entity from the receiver's perspective. Therefore, it is possible to appropriately judge the reliability of received information received from the Internet, etc.

[0083] (Hardware configuration) 13 is a hardware configuration diagram of the information processing device 1. Next, an example of a hardware configuration for realizing each function of the information processing device 1 will be described with reference to FIG.

[0084] 16, the information processing device 1 includes, for example, a CPU (Central Processing Unit) 91, a memory 92, a hard disk 93, and a network interface 94. The CPU 91 is connected to the memory 92, the hard disk 93, and the network interface 94 via a bus.

[0085] The network interface 94 is an interface for communication between the information processing device 1 and an external device. The network interface 94 relays communication between the CPU 91 and, for example, the receiver terminal device 2, the reliability determination model management server 31, the credibility determination tendency model management server 32, the reliability list management server 33, and the graph information management server 34.

[0086] The hard disk 93 is an auxiliary storage device. The hard disk 93 can store the graph information 111, the trust list 112, the trust level determination model 113, and the credibility determination tendency model 114 illustrated in Fig. 1. The hard disk 93 also stores various programs including programs for realizing the functions of the data storage unit 11, the unique graph information generation unit 12, the credibility determination unit 13, the output unit 14, and the machine learning model update unit 15 illustrated in Fig. 1.

[0087] The memory 92 is a main storage device, and may be, for example, a dynamic random access memory (DRAM).

[0088] The CPU 91 reads out various programs from the hard disk 93, and deploys and executes them in the memory 92. In this way, the CPU 91 realizes the functions of the data storage unit 11, the unique graph information generation unit 12, the credibility determination unit 13, the output unit 14, and the machine learning model update unit 15 illustrated in FIG. [Explanation of symbols]

[0089] 1. Information processing device 2. Recipient terminal device 3. Servers 4. Internet 11 Data storage unit 12 Unique graph information generation unit 13 Credibility Assessment Department 14 Output section 15 Machine learning model update section 31 Reliability judgment model management server 32 Credibility judgment tendency model management server 33 Trust List Management Server 34 Graph information management server 35 Social Networking Server 100 Information Processing Systems 111 Graph Information 112 Trust List 113 Reliability Judgment Model 114 Credibility Judgment Tendency Model

Claims

1. a trained first machine learning model that receives property information indicating the certainty of data, the property information being assigned with reliability information indicating the reliability of the property information, and outputs the credibility of the data; acquiring first property information indicating the reliability of the predetermined data received by the receiver; generating first reliability information indicating reliability of the first property information based on a reliability list in which generators of the first property information trusted by the receiver are registered; The first property information to which the first reliability information has been added is input to the first machine learning model to determine the credibility of the predetermined data and present a result. An information processing program that causes a computer to execute a process.

2. the acquisition of the first property information is performed by acquiring graph information that represents a relationship between the predetermined data, the first property information, and the generating subject; The unique graph information obtained by adding the first reliability information to the graph information is used as input data for learning by a first machine learning model.

2. The information processing program according to claim 1,

3. The second machine learning model is trained based on the past reliability judgment results of the receiver for the second property information, and outputs reliability information of the specific property information using specific property information and information of a generator of the specific property information as input. The information processing program according to claim 2, characterized in that the generation of the first reliability information is performed by inputting third property information of the first property information linked to the generator that is not registered in the trust list and information of the generator of the third property information into the second machine learning model to generate the first reliability information.

4. The information processing program according to claim 3, further comprising causing the computer to execute a process of receiving a judgment result of the recipient's reliability for the first property information and updating the second machine learning model.

5. The information processing program according to claim 1, further comprising causing the computer to execute a process of receiving a result of a judgment of the credibility of the recipient for the specified data and updating the first machine learning model based on the result of the judgment of the credibility.

6. The information processing program according to claim 5, characterized in that the computer is caused to execute a process of updating the first machine learning model using virtual property information that is judged to be highly reliable by the receiver in addition to the credibility judgment result.

7. An information processing device, a trained first machine learning model that receives property information indicating the certainty of data, the property information being assigned with reliability information indicating the reliability of the property information, and outputs the credibility of the data; acquiring first property information indicating the reliability of the predetermined data received by the receiver; generating first reliability information indicating reliability of the first property information based on a reliability list in which generators of the first property information trusted by the receiver are registered; The first property information to which the first reliability information has been added is input to the first machine learning model to determine the credibility of the predetermined data and present a result.

2. An information processing method comprising:

8. a data storage unit that acquires first property information indicating the reliability of the predetermined data received by a recipient; a generating unit that generates first reliability information indicating reliability of the first property information based on a reliability list in which generators trusted by the receiver are registered among generators of the first property information acquired by the data holding unit; a credibility determination unit that inputs the first property information to which the first reliability information generated by the generation unit has been assigned into a trained first machine learning model that uses property information indicating the certainty of data and to which reliability information indicating the reliability of the property information has been assigned as an input and outputs the credibility of the data, determines the credibility of the predetermined data, and presents a result; An information processing device comprising:

Images