Decision support system, decision support method, and decision support program

Abstract

To provide users with technology that supports decision-making based on personalized knowledge graphs. [Solution] The system includes a storage unit 120 for storing a hierarchical knowledge graph having multiple prioritized base layers; an interest information acquisition unit 111 for acquiring interest information including the user's interest events; an identification unit 112 for identifying a path to a node in the knowledge graph 121 corresponding to the interest event based on the interest information; a suggestion unit 113 for determining and presenting candidate actions included in the path to the user based on connection coefficients between the nodes; an action acquisition unit 114 for acquiring the selected action chosen by the user from the candidate actions; an update unit 115 for amplifying the connection coefficients between nodes included in the path to the node corresponding to the selected action in the knowledge graph 121; and an output unit 117 for outputting the amplified connection coefficients between nodes included in the path to the node corresponding to the selected action to the user.

Description

Technical Field

[0001] The present invention relates to a system, method, and program for assisting decision-making based on a personalized knowledge graph for a user.

Background Art

[0002] Conventionally, technologies related to digital clones that reproduce an individual's thinking pattern and behavioral characteristics in a virtual space are known. Such digital clones are intended to substitute for the decision-making of the individual or simulate the individual's reaction under specific circumstances, and are expected to be utilized in a wide range of fields such as marketing, counseling, or inheritance of an individual's will.

[0003] In order to improve the accuracy of digital clones, an important issue is how to accurately extract and learn the idiosyncrasies of thinking and values unique to the individual from a variety of data of the target individual.

[0004] Patent Document 1 discloses an information processing technology for constructing a human digital twin. In this technology, the responses given by the monitored member himself / herself collected via a questionnaire server are used as learning data, and adjustment (tuning) of the digital twin is performed. Specifically, the responses of the individual and the digital twin to specific questions are compared, and the digital twin is optimized based on the difference, thereby reproducing the thinking of the individual more faithfully.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] Typical digital clones are constructed by providing a machine learning model with the individual's personal data as training data. However, the internal structure of such digital clones is either a black box or a complex structure incomprehensible to humans, making it practically difficult to verify whether they accurately reflect the individual's thought patterns and behavioral characteristics.

[0007] In light of the circumstances described herein, this disclosure aims to address the problem of providing users with technology to support their decision-making based on a personalized knowledge graph. [Means for solving the problem]

[0008] To solve the above problems, the decision support system described herein is a decision support system that assists users in making decisions based on a personalized knowledge graph, A memory unit that stores a hierarchical knowledge graph having multiple prioritized base layers, A unit for acquiring interest information that acquires interest information including the user's areas of interest, A selection unit that identifies a path to the node corresponding to the interest event in the knowledge graph based on the interest information, A proposal unit that determines the candidate actions included in the aforementioned path based on the connection coefficient between the nodes and presents them to the user, An action acquisition unit that acquires the selected action chosen by the user from the aforementioned action candidates, An update unit that amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the selection action in the knowledge graph, The system includes an output unit that outputs to the user an amplified connection coefficient between nodes included in the path leading to the node corresponding to the selection action. [Effects of the Invention]

[0009] According to this disclosure, it is possible to provide technology that supports users in making decisions based on a personalized knowledge graph. [Brief explanation of the drawing]

[0010] [Figure 1] A schematic diagram of the decision support system of this embodiment. [Figure 2] Hardware configuration diagram of the generation device according to this embodiment. [Figure 3] Hardware configuration diagram of the user terminal in this embodiment. [Figure 4] Hardware configuration diagram of the wearable device according to this embodiment. [Figure 5] Block diagram of the generation apparatus of this embodiment. [Figure 6] A schematic diagram illustrating the configuration of the knowledge graph in this embodiment. [Figure 7] A processing flowchart of the generation method according to this embodiment. [Figure 8] An example of how the knowledge graph of this embodiment is displayed. [Figure 9] A block diagram of the inference device of this embodiment. [Figure 10] A processing flowchart of the inference method of this embodiment. [Figure 11] A conceptual diagram of the knowledge graph in this embodiment. [Modes for carrying out the invention]

[0011] The following description of the decision support system, method, and program relating to embodiments of this disclosure will be illustrated with reference to the drawings. Note that the embodiments shown below are examples of the present invention, and this disclosure is not limited to these embodiments; various configurations can be adopted.

[0012] In this embodiment, the configuration, operation, etc. of the decision-making support system will be described. However, a method with a similar configuration, a computer program, and a program recording medium on which the program is recorded, etc. also exhibit similar effects. By using the program recording medium, the program can be installed in a computer. A series of processes according to this embodiment described below are provided as a program executable by a computer and can be provided via non-transitory computer-readable recording media such as CD-ROMs and flexible disks, and further via communication lines.

[0013] The decision-making support system is composed of one or more computer devices. The computer device includes an arithmetic device such as a CPU (Central Processing Unit) and a storage device. By executing the program stored in the storage device by the arithmetic device, the computer device can function as a generation device or an inference device. The decision-making support method is realized by the processing of each device including the generation device and the inference device.

[0014] 1. System Configuration FIG. 1 shows a schematic diagram of the decision-making support system 1. The decision-making support system 1 includes a generation device 100, an inference device 200, a user terminal 300, and a wearable device 400. The generation device 100 is connected via a communication network NW to the user terminal 300. The user terminal 300 is communicatively connected to the wearable device 400 by wire or wirelessly. The decision-making support system 1 executes a decision-making support method as a whole through the cooperation of each method executed by each component. In the illustrated example, only one user terminal 300 and one wearable device 400 are shown, but there may be a plurality of these. Also, the wearable device 400 may be connected via the communication network NW to the generation device 100 without going through the user terminal 300.

[0015] The generation device 100 generates a knowledge graph 121 based on the generation data acquired from the user terminal 300. The generation device 100 can update the knowledge graph 121 based on the generation data intermittently acquired from the user terminal 300. The inference device 200 acquires the knowledge graph 121 generated by the generation device 100 and has it as the knowledge graph 222. The inference device 200 can acquire inference target data from an external device and output an inference result based on the knowledge graph 222. The generation device 100 and the inference device 200 may be communicatively connected by wire or wirelessly. Also, the inference device 200 may be configured to acquire inference target data and output an inference result based on the knowledge graph 121 possessed by the generation device 100.

[0016] The inference device 200 acquires the knowledge graph 121 updated by the generation device 100 in real time or based on a predetermined trigger. The predetermined trigger includes the elapse of a predetermined period, the update of the knowledge graph 121, and the like. The inference device 200 can synchronize (update) the knowledge graph 222 based on the acquired knowledge graph 121.

[0017] 2. Hardware Configuration FIG. 2 shows a hardware configuration diagram of the generation device 100. The generation device 100 is constituted by a server device, a personal computer, or the like, and includes a processor 110, a memory 120, and a communication interface 130. The generation device 100 may further include an input / output interface (such as a keyboard and a monitor). The generation device 100 may be constituted by one or a plurality of computers.

[0018] The processor 110 consists of a CPU or GPU (Graphics Processing Unit) capable of executing instruction sets. The processor 110 controls the overall operation of the generation device 100 by executing a computer program that executes the generation method, or at least a portion of multiple computer programs that execute collaboratively, as well as an OS (Operating System) and other applications. The generation device 100 includes one or more processors 110.

[0019] The memory 120 includes volatile memory such as RAM (Random Access Memory) capable of storing instruction sets, and non-volatile recording media such as HDD (Hard Disk Drive) or SSD (Solid State Drive) capable of storing the OS and the aforementioned computer programs. The generation device 100 includes one or more memories 120, with at least one of the one or more memories storing the knowledge graph 121 and at least one storing a database management system that functions as a database.

[0020] The communication interface 130 has an interface for connecting to a communication network and performs communication control to input and output information. The generation device 100 includes one or more communication interfaces 130. If the generation device 100 is operated as a server and does not have a physical input / output interface, it is operated by a client via a communication network such as a LAN (Local Area Network) or WAN (Wide Area Network). Similarly, when multiple generation devices 100 work together to execute a decision support method, communication between the multiple generation devices 100 is performed via the communication network. Furthermore, when using a general-purpose machine learning model provided by an external service provider via an API, the generation device 100 can also connect to the WAN via the communication interface 130 and use the machine learning model.

[0021] The inference device 200 has the same hardware configuration as the generation device 100. The processor of the inference device 200 controls the overall processing of the inference device 200 by executing a program that executes the inference method stored in memory, an OS, and other applications. The inference device 200 may also be configured as an autonomous robot. The autonomous robot includes a processor, memory, a communication interface, sensors, actuators, etc. Multiple actuators are provided corresponding to the arms and legs of the autonomous robot, enabling movement in the physical environment and physical interaction with objects. The autonomous robot may be configured independently of the inference device 200 and operate by acquiring the inference results (control signals, etc.) from the inference device 200.

[0022] Figure 3 shows the hardware configuration diagram of the user terminal 300. The user terminal 300 is composed of a personal computer, tablet terminal, or smartphone, and includes a processor 310 such as a CPU, memory 320 such as RAM, HDD, SSD, or flash storage, a communication interface 330 for connecting to a communication network via wireless communication or wired LAN, an input interface 340 such as a mouse, keyboard, touch panel, or microphone, and an output interface 350 such as a display or speaker. Multiple user terminals 300 may exist, and they are operated by the user who generates the knowledge graph 121. The memory 320 stores browser applications and various data. The processor 310, for example, executes a browser application and accesses web services provided by the generation device 100 via a web browser, thereby enabling data communication with the generation device 100.

[0023] Figure 4 shows the hardware configuration diagram of the wearable device 400. The wearable device 400 consists of smart glasses or AR (Augmented Reality) glasses worn by the user and includes a processor 410 such as a CPU or GPU, memory 420 such as RAM, HDD, SSD, or flash storage, a communication interface 430 that connects to at least the user terminal 300 by short-range wireless communication, wireless, or wired, a sensor 440, a microphone 450, and a display 460.

[0024] Sensor 440 is configured as at least one sensor selected from a camera, accelerometer, angular velocity sensor, tracking sensor, etc. The camera captures images or videos of the real space that constitutes the user's field of view. The accelerometer and angular velocity sensor detect the position and orientation of the wearable device 400. The tracking sensor detects the movement of the user's face, eyes, etc. The tracking sensor can be configured, for example, as an eye-tracking sensor that detects the movement of the user's pupils, and can detect data such as the object the user is looking at and the duration of that attention.

[0025] The microphone 450 detects the voice emitted by the user. The processor 410 can transmit the voice data detected by the microphone 450, or the text data of said voice, to an external device.

[0026] Display 460 is a transparent display that overlays augmented reality graphics onto the real space included in the user's field of view.

[0027] 3. Functional configuration of the generating device Figure 5 shows a block diagram of the generation device 100. The processor 110 of the generation device 100 functions as a functional component including an interest information acquisition unit 111, an identification unit 112, a proposal unit 113, an action acquisition unit 114, an update unit 115, a knowledge analysis unit 116, and an output unit 117 by executing a computer program stored in the memory (storage unit) 120. The memory 120 stores data including a knowledge graph 121 and a reference connection coefficient 122.

[0028] The generation device 100 has one or more processors, at least one of which functions as an interest information acquisition unit 111, at least one as a specification unit 112, at least one as a suggestion unit 113, at least one as an action acquisition unit 114, at least one as an update unit 115, at least one as a knowledge analysis unit 116, and at least one as an output unit 117. Some or all of the above-described functional components may be executed by a processor in the inference device 200, user terminal 300, or wearable device 400.

[0029] The knowledge graph 121 is stored in memory 120, linked to user identification information for identifying the user. The knowledge graph 121 consists of multiple nodes and edges connecting the nodes. The knowledge graph 121 has a hierarchical structure including a base layer composed of multiple layers. The multiple layers of the base layer are prioritized, and if the criteria of a lower layer are not met, referencing the criteria of a higher layer or processing based on those criteria is restricted. The base layer can, but is not limited to, a needs system, an economic rationality system, or a time axis system.

[0030] The needs system is a framework that represents five needs, in descending order from the lowest level: physiological needs, safety needs, social needs, esteem needs, and self-actualization needs. Knowledge graph 121, which uses this as its base layer, assigns high priority to the levels of physiological needs and safety needs. The economic rationality system is a framework that represents four levels of economic rationality, from the lowest level upwards: principal preservation, securing survival costs, default risk, and profit maximization. Knowledge graph 121, which uses this as its base layer, assigns high priority to levels related to economic security, such as principal preservation and securing survival costs. The time axis system represents four time axes, in descending order from the lowest level: today's survival, short-term survival, medium-term survival, and long-term survival. Knowledge graph 121, which uses this as its base layer, assigns higher priority to the shorter-term safety level.

[0031] In this embodiment, we describe an example in which the knowledge graph 121 employs a hierarchical structure with multiple stages of needs as its base layer. The base layer is classified into nodes representing five needs: physiological needs, safety needs, social needs, esteem needs, and self-actualization needs. The base layer has an event layer below it, and each node in the base layer is connected to multiple nodes in the event layer by edges.

[0032] The event layer consists of one or more layers and is composed of nodes representing multiple events. For example, the nodes of the first layer of the event layer are connected by edges to one or more nodes of the second layer, and the nodes of the second layer are further connected by edges to one or more nodes of the third layer. The event layer may have a different number of layers for each event, and there are no restrictions on the type of event or the number of layers. The event layer has nodes that represent more specific events as you go down the layers. The event layer has an action layer below it, and each node of the event layer can be connected by edges to multiple nodes of the action layer.

[0033] The behavior layer consists of a single hierarchy and is composed of nodes representing multiple actions. Nodes in the behavior layer may be connected by edges to nodes in any hierarchy of the higher-level event layer. For example, nodes in the behavior layer can be connected by edges not only to the lowest-level nodes of the event layer, but also to nodes in the layers above it.

[0034] The edges connecting each node have a connection coefficient that represents the strength of the connection between the nodes. The connection coefficient is set as a numerical range, for example, from 0 to 1, where a value closer to 0 indicates a weaker connection between nodes, and a value closer to 1 indicates a stronger connection between nodes.

[0035] Figure 6 shows an overview of the structure of knowledge graph 121. Knowledge graph 121 has nodes N1 to N5 in the base layer. In the illustrated example, knowledge graph 121 displays only some nodes in the event layer and behavior layer that are connected to node N2, which represents the need for security in the base layer. Node N2 is connected to node N21, which represents property in the first layer of the event layer. Node N21 is connected to node N22, which represents investment in the second layer of the event layer. Node N22 is connected to node N23, which represents stocks in the third layer of the event layer. Node N23 is connected to node N24, which represents the first stock, and node N25, which represents the second stock, in the fourth layer of the event layer. In the event layer, the events become more concrete as you go from the top to the bottom layers, from the event of property to the specific stock. Thus, the knowledge graph 121 shown in Figure 6 unfolds from the higher levels of abstraction to the lower levels of concreteness. In the event layer, the edges from parent nodes to child nodes represent the subdivision of events, while the edges from the event layer to the action layer represent the feasibility of taking specific actions related to that event.

[0036] In Figure 6, node N24 is connected to node N201, which represents the latest news in the action layer; node N202, which represents a comparison with other companies in the same industry; and node N203, which represents a buy or sell order. Each node in the action layer represents a specific action related to the event represented by the node in the connected event layer. Each node in the action layer functions as a trigger to cause the system to execute the action represented by the node when the proposal unit 113 or the decision-making unit 213, described later, identifies that node. For example, node N201 represents the action of acquiring and presenting the latest news regarding the first stock, node N202 represents the action of acquiring and presenting the comparison results between the first stock and other stocks in the same industry, and node N203 represents the action of placing a buy or sell order for the first stock.

[0037] First, memory 120 stores the initially configured knowledge graph 121. Based on generation data acquired for each user by the user terminal 300, the knowledge graph 121 is updated to include the connection relationships and connection coefficients of nodes and edges included in the event layer and behavior layer, thereby generating a knowledge graph 121 personalized for the user.

[0038] The interest information acquisition unit 111 acquires interest information, including the user's interests. Specifically, the interest information acquisition unit 111 identifies the user's interests based on the data acquired from the user terminal 300. The data acquired from the user terminal 300 includes data such as text, images, and audio input via the input interface 340 of the user terminal 300, as well as sensor data detected by the sensor 440 of the wearable device 400 and audio data input via the microphone 450.

[0039] The interest information acquisition unit 111 can acquire interest information based on the user's gaze information detected by the wearable device 400. Specifically, the interest information acquisition unit 111 identifies interest events based on image information detected by the camera and gaze information detected by the tracking sensor. The image information represents an image of the real space corresponding to the user's field of view. The gaze information represents information about the user's gaze movement relative to the image information. Based on the image information and gaze information, the interest information acquisition unit 111 can analyze the object that the user is focusing on in the image information (real space) from the movement of their gaze and the time they are focusing on it, and identify it as an interest event.

[0040] The identification unit 112 identifies a path to the node corresponding to the event of interest in the knowledge graph 121 based on the information of interest. The event of interest is identified from among the events represented by the nodes included in the event layer of the knowledge graph 121. If the identification unit 112 cannot identify a node corresponding to the event of interest, it may generate a new node corresponding to the event of interest.

[0041] The suggestion unit 113 determines the action candidates included in the path to the identified node based on the connection coefficients between nodes and presents these action candidates to the user. Specifically, the suggestion unit 113 displays one or more action candidates via the output interface 350 of the user terminal 300 or the display 460 of the wearable device 400, prompting the user to select an action candidate. The path to the identified node is the path from the base layer node to the identified node in the event layer. This path includes multiple event layer nodes and edges. The action candidates are determined from among the action layer nodes connected to the event layer nodes included in these paths. Note that the action candidates include action layer nodes connected to the identified node.

[0042] The proposal unit 113 determines action candidates based on the connection coefficients of the edges connecting the event layer nodes and action layer nodes included in the path. Specifically, the proposal unit 113 extracts all action layer nodes that could be action candidates and filters them in descending order of the connection coefficients of the edges connected to those action layer nodes. The proposal unit 113 determines one or more action layer nodes whose connection coefficients are above a predetermined threshold as action candidates. If the proposal unit 113 is limited in the number of action candidates that can be presented to the user (e.g., three or fewer), it determines the action candidates in descending order of connection coefficients according to that limit. If the proposal unit 113 is limited in the number of action candidates that should be presented to the user (e.g., two or more), it may determine the action candidates in descending order of connection coefficients according to that limit, regardless of the connection coefficient threshold.

[0043] The proposal unit 113 can generate new action candidates not included in the knowledge graph 121 based on past connection coefficient trends in the knowledge graph 121 and present these action candidates to the user. The proposal unit 113 can refer to nodes of events similar to the events of the event layer nodes connected to the action candidate nodes and generate new action candidates based on the actions corresponding to the action layer nodes connected to the nodes of the similar events. Alternatively, the proposal unit 113 may input events corresponding to the event layer nodes connected to the action candidate nodes into an inference model and generate action candidates based on the output from the inference model. The inference model can be a large-scale language model or the like. Furthermore, the proposal unit 113 may refer to the knowledge graphs of other users and generate new action candidates based on the actions corresponding to the action layer nodes connected to nodes representing the same events. The proposal unit 113 may generate new action candidates and also generate action layer nodes corresponding to those action candidates.

[0044] The action acquisition unit 114 acquires the selected action chosen by the user from among the action candidates. Specifically, the action acquisition unit 114 receives the user's selection from one or more action candidates displayed via the output interface 350 of the user terminal 300 or the display 460 of the wearable device 400, via the user terminal 300 or the wearable device 400. In the wearable device 400, the selection of an action candidate is accepted by the user's gaze toward the action candidate displayed on the display 460 or by voice input via the microphone 450. The action acquisition unit 114 may also acquire the user's selected action that is not included in the action candidates presented by the suggestion unit 113.

[0045] The update unit 115 amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the selected action in the knowledge graph 121. The update unit 115 also attenuates the connection coefficients between nodes included in the path leading to the node corresponding to the action candidate that was not selected as the selected action in the knowledge graph 121. The update unit 115 stores the knowledge graph 121, which reflects the updated (amplified or attenuated) connection coefficients, in the memory 120. The targets for amplification or attenuation of connection coefficients include all edges from the base layer node to the action layer node corresponding to the selected action candidate (selected action) or the action candidate that was not selected. Note that the weights of the connection coefficients amplified or attenuated may differ for each layer.

[0046] When the update unit 115 acquires an action candidate that is not included in the knowledge graph 121 as a selected action, it generates a new action layer node corresponding to the action candidate included in that selected action, adds it to the knowledge graph, and amplifies the connection coefficients between nodes included in the path leading to that node.

[0047] The update unit 115 stores the update history of connection coefficients due to selection behavior in memory 120, linking it to the knowledge graph 121. The update history includes identification information of the updated edge, the type of update (amplification or attenuation), and information indicating the time of the update. The update history may also include the strength of the updated connection coefficient and identification information of the node corresponding to the updated edge. The information indicating the time represents the year, month, and day or date and time when the connection coefficient was updated.

[0048] The output unit 117 outputs to the user information regarding the amplification of connection coefficients between nodes included in the path leading to the node corresponding to the selected action. The output unit 117 provides feedback in an output format that allows the user to perceive the fact that the connection coefficients have been amplified. The amplification information is output, for example, as visual, auditory, or tactile feedback, or as a control signal to an external device.

[0049] As a form of visual feedback, the output unit 117 presents the amplification of connection coefficients between nodes included in the path leading to the node corresponding to the selected action to the user through visual changes in the graphical user interface (GUI). Specifically, the output unit 117 presents the entire knowledge graph 121 to the user as a GUI, and when an update occurs due to amplification of connection coefficients in conjunction with a selected action, it displays the amplification of connection coefficients in the path from the action layer node corresponding to the selected action to the base layer node through dynamic visual changes. Visual changes include illumination or color changes of the entire path, and animations representing the reverse flow of energy from the action layer node to the base layer node. Note that the visual changes are not limited to these examples, and any representation that allows the user to visually recognize the amplification of connection coefficients can be adopted.

[0050] The output unit 117 can present the knowledge graph 121 to the user via the output interface 305 of the user terminal 300 or the display 460 of the wearable device 400. When outputting to the display 460, the output unit 117 processes the knowledge graph 121 to display as a virtual object superimposed on real space.

[0051] The output unit 117 highlights paths in the knowledge graph 121 on the GUI as the connection coefficient between nodes increases, and does not highlight paths in the knowledge graph 121 on the GUI as the connection coefficient between nodes decreases. The output unit 117 may also hide paths if the connection coefficient falls below a predetermined threshold.

[0052] As a form of auditory feedback, the output unit 117 amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the chosen action and outputs the changes in the connection coefficients, the accompanying advice, warnings, a score regarding the validity of the current decision (such as the strength of the connection coefficients), or a combination thereof, as synthesized speech via a speaker configured as the output interface 350 of the user terminal 300 or a speaker mounted on the wearable device 400.

[0053] As a form of tactile (physical) feedback, the output unit 117 outputs an amplification of the connection coefficients between nodes included in the path leading to the node corresponding to the selected action as vibrations of a haptic device mounted on the wearable device 400 worn by the user. Alternatively, the output unit 117 may output control signals to provide physical feedback via the actuators of an autonomous robot. Physical feedback may include, for example, a form in which the robot physically restrains a user attempting to select a specific action.

[0054] Furthermore, as a form of feedback to external devices via control signals, the output unit 117 outputs an amplification of the connection coefficients between nodes included in the path leading to the node corresponding to the selected action, as a control signal for an external system present in the user's surrounding environment. The external system includes web services that the user terminal 300 accesses simultaneously and IoT devices synchronously connected to the user terminal 300. The control signals include restricting or stopping the use of external systems, as well as controlling the operating mode of IoT devices, providing direct or indirect feedback to the user.

[0055] In this way, the amplification of the connection coefficient is perceptually fed back to the user, allowing the user to immediately recognize the reflection of their own thought patterns and behavioral characteristics in the knowledge graph 121. Furthermore, they can objectively analyze their own thinking by linking their thought patterns and behavioral characteristics to their fundamental desires.

[0056] The knowledge analysis unit 116 evaluates the consistency of the update history of connection coefficients associated with the selection behavior of a predetermined topic over a predetermined period. The predetermined period is set to a period of at least one month (for example, 6 months, 1 year, 3 years, 5 years, etc.). In this embodiment, 5 years is set, but it is not limited to this and any period can be set.

[0057] A predetermined selection action includes actions represented by nodes in the action layer related to a predetermined topic. A topic is an identification index for defining a partial network (hereinafter referred to as a topic graph) that has a common context or purpose among multiple nodes and edges that constitute the knowledge graph 121. A topic is defined as a set of paths starting from each node N1 to N5 in the base layer, or as a hierarchical structure with a specific node in a specific layer of the event layer as the root. For example, by designating node N21 representing property or node N22 representing investment in Figure 6 as specific nodes, the network under them is defined as a topic graph. A topic is defined by selecting any node by the user or system administrator. Alternatively, a topic may be automatically defined by the knowledge analysis unit 116 according to the update frequency of connection coefficients for specific paths and nodes over a predetermined period.

[0058] The knowledge analysis unit 116 evaluates the consistency of the update history of connection coefficients according to at least one consistency index selected from the variance of connection coefficients, the selection frequency of a specific node, and the reaction time of the selection behavior. When the knowledge analysis unit 116 evaluates consistency by combining multiple consistency indexes, it integrates the consistency indexes by weighting each index or the like, and sets the period during which the consistency index exceeds a predetermined threshold as the reference period.

[0059] The variance of the connection coefficient is calculated as the variance of the connection coefficient for a specific path during the target period. The knowledge analysis unit 116 highly evaluates the consistency index, considering the user's decision-making to be consistent, if the variance is below a predetermined threshold. The frequency of selecting a particular node is calculated as the probability of selecting a node representing a specific action that is connected to a node representing a certain event. The knowledge analysis unit 116 highly evaluates the consistency index, considering the user's decision-making to be consistent, if the probability of selecting a node representing a specific action is above a predetermined threshold. The reaction time for a selected action is calculated as the time from when the suggestion unit 113 presents action candidates to the user until the action acquisition unit 114 acquires the selected action from the user. If the reaction time is below a predetermined threshold, the knowledge analysis unit 116 evaluates the consistency index highly, considering the user's decision-making to be consistent.

[0060] The knowledge analysis unit 116 stores the connection coefficients between nodes included in a consistent topic in memory 120 as a reference connection coefficient 122, linking them to a time attribute. The reference connection coefficient 122 can be the mean, median, or representative value of the connection coefficients during the reference period. The time attribute is a label representing the reference period of the update history to be evaluated for consistency, and is generated as a label representing the period from the earliest update history date and time to the latest update history date and time. The time attribute may also be generated as a label representing the user's age during a predetermined period.

[0061] The base connection coefficient 122 includes a topic, connection coefficients between nodes included in that topic, and time attributes. In other words, the base connection coefficient 122 can have independent connection coefficients and time attributes for each topic. This makes it possible to optimize by age on a topic-by-topic basis, for example, by identifying a base connection coefficient based on the update history of the time attribute "30s" for the topic "Investment," and by identifying a base connection coefficient based on the update history of the time attribute "40s" for the topic "Health Management."

[0062] 4. Generation method The decision support method of this disclosure includes a generation method for generating a knowledge graph personalized for a user, and an inference method for performing reasoning related to decision-making using the knowledge graph.

[0063] Figure 7 shows the processing flowchart of the generation method. The generation method is realized by the functional components of the generation device 100.

[0064] In step S101, the generation device 100 acquires interest information, including the user's interests, from the user terminal 300 or wearable device 400. In step S102, the generation device 100 identifies a path to a node corresponding to an event of interest in the knowledge graph based on the acquired interest information. The generation device 100 may also provide a GUI to the user terminal 300 or wearable device 400 that displays the identified node and the path to that node on the knowledge graph. In step S103, the generation device 100 determines the candidate actions included in the path to the identified node based on the connection coefficient between the nodes, and presents one or more of the determined candidate actions to the user terminal 300 or wearable device 400. In step S104, the generation device 100 obtains the selection action selected by the user from the user terminal 300. In step S105, the generation device 100 amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the selected action in the knowledge graph. The generation device 100 also attenuates the connection coefficients between nodes included in the path leading to the node corresponding to the action candidate that was not selected as the selected action in the knowledge graph. The generation device 100 provides the amplified connection coefficients between nodes included in the path leading to the node corresponding to the selected action to the user terminal 300 or wearable device 400 through visual changes in the GUI.

[0065] Figure 8 shows an example of how the knowledge graph 121 presented to the user at each step of the generation method is displayed. This knowledge graph 121 may be displayed as a virtual object via the display 460 of the wearable device 400, or it may be displayed on the display of the user terminal 300.

[0066] Figure 8(a) shows an example of a display of the knowledge graph 121 relating to the paths leading to the nodes corresponding to the events of interest identified in process S102. In the example, the user's event of interest is the second brand, and the paths from nodes N2, N21, N22, N23, and N25 leading to node N25, which corresponds to the second brand, are displayed in a way that distinguishes them from other paths. The display methods can include, for example, highlighting, illumination, and color changes.

[0067] Figure 8(b) shows an example of the display of the knowledge graph 121 regarding the action candidates presented in process S103. In the example, nodes N201, N202, and N203, which correspond to the action candidates connected to node N25, which represents a second stock corresponding to the identified event, are displayed. In addition, node N201, which represents recent news, is highlighted as the most appropriate action candidate based on the user's decision-making tendencies. Note that the display 460 may display only the single optimal action candidate selected from multiple action candidates (node ​​N201 in the example).

[0068] Figure 8(c) shows an example of a display of a knowledge graph 121 relating to visual changes that represent the amplification of connection coefficients between nodes included in the path leading to the node corresponding to the user's chosen action in process S105. In the example shown, node N202, representing comparison within the same industry, is acquired as the user's chosen action, and the amplification of connection coefficients between nodes in the path from node N202 to nodes N25, N23, N23, N22, N21, and N2 is represented by a visual change through animation caused by the reverse flow of energy from node N202. On the other hand, the connection coefficients between nodes N201 and N203, which correspond to action candidates that were not selected as the chosen action, and node N25 to which these nodes are connected, are attenuated, and the path becomes narrower. In this way, by showing the results of amplification or attenuation of connection coefficients in the path leading to the node representing the action as differences in the display manner, a knowledge graph 121 that represents the user's decision-making and thinking habits can be generated.

[0069] 5. Functional Configuration of the Inference System Figure 9 shows a block diagram of the inference device 200. The processor 210 of the inference device 200 functions as a functional component including a reference information acquisition unit 211, an interest information acquisition unit 212, a decision-making unit 213, an arbitration unit 214, and a control signal output unit 215 by executing a computer program stored in the memory (storage unit) 220. The processor 210 of the inference device 200 may also utilize the functional components of the knowledge analysis unit 116 of the generation device 100. The memory 220 stores data including reference information 221, a knowledge graph 222, and reference connection coefficients 223. Note that the knowledge graph 222 and reference connection coefficients 223 include data generated by the generation device 100.

[0070] The inference device 200 has one or more processors, at least one of which functions as a reference information acquisition unit 211, at least one as an interest information acquisition unit 212, at least one as a decision-making unit 213, at least one as an arbitration unit 214, and at least one as a control signal output unit 215. Some or all of the above-described functional components may be executed by a processor in any of the generation device 100, the user terminal 300, or the wearable device 400.

[0071] The inference device 200 may be used in parallel with the generation of the knowledge graph by the generator 100. The inference device 200 can acquire the knowledge graph 121 generated by the generator 100 through dynamic or periodic updates and use it as the knowledge graph 222 for inference.

[0072] The reference information acquisition unit 211 acquires reference information of the knowledge graph 222 to be referenced in decision-making from the user. The reference information includes time attributes and topics. The topic of the reference information refers to a partial network of the knowledge graph 222 defined as a topic graph. The reference information acquisition unit 211 can acquire multiple reference information 221, each containing different time attributes for each topic, and store them in the memory 220. For example, it can acquire reference information associated with a first time attribute for a first topic and a second time attribute for a second topic. This allows the decision-making unit 213, described later, to selectively reference different time-based reference connection coefficients 223 for each topic.

[0073] The interest information acquisition unit 212 acquires interest information, including the user's interest events, similar to the interest information acquisition unit 111 of the generation device 100.

[0074] The decision-making unit 213 provides the user with information to support decision-making based on the knowledge graph 222. Specifically, based on the interest information, the decision-making unit 213 can identify a path to a node corresponding to an event of interest in the knowledge graph 222, and propose candidate actions included in that path based on the strength of the connection coefficients in the knowledge graph 222.

[0075] The decision-making unit 213 prioritizes proposing action candidates corresponding to nodes at the end of paths (action layer) with high connection coefficient strengths. Specifically, it proposes action candidates corresponding to the action layer node where the sum of the connection coefficient strengths included in the path from the node corresponding to the event of interest to the action layer node is maximized. The node corresponding to the event of interest is not limited to nodes directly connected to the action layer node, but may also be a node in a higher layer of the event layer. For example, even if the event of interest is a node in a higher layer, the system can proactively identify the action layer node that the user is most likely to ultimately select, without waiting to identify the event in the lower layer.

[0076] The decision-making unit 213 can adjust the weights of the connection coefficients according to the prioritized hierarchy of the base layer. For example, if the strength of the connection coefficients of a node in a lower base layer and a node in a higher base layer are similar, a greater weight is assigned to the connection coefficient of the node in the lower base layer. This allows the decision-making unit 213 to propose action candidates corresponding to the node in the lower base layer with greater priority.

[0077] The decision-making unit 213 provides the user with information to support decision-making based on the reference connection coefficients. Specifically, the decision-making unit 213 identifies topics corresponding to the events of interest included in the interest information obtained from the user. The decision-making unit 213 identifies the time attributes associated with the identified topic in the reference information containing that topic. The decision-making unit 213 obtains the reference connection coefficients 223 associated with the identified topic and time attributes from the memory 220. The decision-making unit 213 can determine candidate actions based on the connection coefficients included in the obtained reference connection coefficients 223.

[0078] If the decision-making unit 213 does not have reference information containing the topic identified based on the interest information, it retrieves a reference connection coefficient 223 associated with that topic from the memory 220. The decision-making unit 213 can determine a candidate action based on the connection coefficients included in the retrieved reference connection coefficient 223. If the decision-making unit 213 retrieves multiple reference connection coefficients 223 associated with a topic from the memory 220, it passes the information of those reference connection coefficients 223 to the mediation unit 214.

[0079] The mediation unit 214, when multiple reference connection coefficients 223 related to topics corresponding to interest information obtained from the user exist and are associated with multiple time attributes, executes a mediation process in which the mediation agent determines the priority of multiple reference connection coefficients 223 (time attributes) through dialogue with the user. The dialogue between the mediation agent and the user is conducted via a GUI using the output interface 350 of the user terminal 300 or the display 460 of the wearable device 400. This dialogue may also be conducted via voice input / output between the mediation agent and the user. The mediation agent can employ large-scale language models or general artificial intelligence. Specifically, the mediation agent prompts the user to select which of the multiple reference connection coefficients 223 to adopt. At this time, the mediation agent can also present the time attributes associated with each reference connection coefficient 223. The mediation agent may also propose to the user the reference connection coefficient 223 that is statistically most consistent in the past, based on the average strength or sum of strengths of each reference connection coefficient 223. The user selects one standard connection coefficient 223 to adopt, based on the information presented by the mediation agent. The mediation unit 214 determines the priority of multiple standard connection coefficients 223 through dialogue with the user. For example, the mediation unit 214 may decide to adopt the standard connection coefficient selected by the user as having the highest priority. The decision-making unit 213 can determine candidate actions based on the connection coefficients included in the standard connection coefficient 223 determined according to the mediation result.

[0080] The control signal output unit 215 outputs a control signal to suppress the user's chosen action if the action candidates presented by the decision-making unit 213 conflict with the user's chosen action. The control signal to suppress the chosen action includes signals to display a warning via the output interface 350 of the user terminal 300 or the GUI of the display 460 of the wearable device 400, signals to restrict user operations on the GUI, and signals to drive actuators to suppress physical user actions by the autonomous robot.

[0081] The control signal output unit 215 may have a restraining criterion for determining whether the proposed action and the selected action are contradictory. The restraining criterion may be at least one criterion or a combination thereof, such as: a selected action different from the proposed action; the connection coefficient of the selected action with respect to the connection coefficient of the proposed action is below a threshold; or the proposed action presented based on the reference connection coefficient is different from the selected action based on the current connection coefficient.

[0082] The current connection coefficients of candidate action nodes related to a particular topic in the knowledge graph 222 will have different strengths than the past connection coefficients included in the baseline connection coefficients of candidate action nodes related to the same topic. In this case, since the baseline connection coefficients are generated based on the user's own reliable past thoughts and actions, it is desirable to prioritize the use of past connection coefficients. For example, suppose that in the topic "investment," the node "investment ratio within 3%" in the past action layer has the highest connection coefficient and is stored in memory 220 as the baseline connection coefficient 223. On the other hand, also in the topic "investment," the node "investment ratio within 3%" in the current action layer has a weaker connection coefficient than the nodes in other action layers, and the knowledge graph 222 based on this connection coefficient is stored in memory 220. In this case, the control signal output unit 215 can determine that if the "investment action with an investment ratio of 20%" is to be executed as the current selected action, it contradicts the node "investment ratio within 3%" in the past action layer in the baseline connection coefficient 223. This allows for the dynamic correction of decision-making errors caused by current emotional fluctuations or inappropriate external influences through consistent self-discipline from the past.

[0083] 6. Reasoning method Figure 10 shows a processing flowchart of the inference method. The inference method is implemented by the functional configuration of the inference device 200.

[0084] In step S201, the inference device 200 acquires reference information via a user terminal 300 or a wearable device 400. In step S202, the inference device 200 acquires interest information, including the user's interest events, via the user terminal 300 or wearable device 400. In step S203, the inference device 200 determines whether a topic corresponding to the event of interest included in the information of interest is defined. If there is no corresponding topic (NO in step S203), in step S204, the inference device 200 identifies a path to the node corresponding to the event of interest, determines candidate actions included in that path based on the connection coefficients of the knowledge graph 222, and presents them to the user. If a corresponding topic exists (YES in step S203), in step S205, the inference device 200 determines whether reference information containing that topic is stored in memory 220. If a corresponding reference information exists (YES in step S205), in step S206, the inference device 200 identifies the time attribute associated with the topic in that reference information and determines a candidate action based on the identified topic and the reference connection coefficient 223 associated with the time attribute. If no corresponding reference information exists (NO in step S205), in step S207, the inference device 200 determines whether there are multiple reference connection coefficients 223 (time attributes) corresponding to the topic identified in step S203. If there are not multiple corresponding reference connection coefficients 223 (NO in step S207), the inference device 200 proceeds to step S206 and determines a candidate action based on the reference connection coefficients 223 associated with the identified topic and time attributes. If there are multiple corresponding reference connection coefficients 223 (time attribute) (YES in step S207), in step S208, the inference device 200 performs arbitration processing. The inference device 200 determines the priority of the multiple reference connection coefficients 223 through dialogue between the arbitration agent and the user. In step S209, the inference device 200 determines a candidate action based on the reference connection coefficient 223 determined according to the mediation result.

[0085] Figure 11 is a conceptual diagram of the knowledge graph 222 in the inference device 200. Figure 11(a) is a conceptual diagram of the knowledge graph 222 based on the current connection coefficient, and Figure 11(b) is a conceptual diagram of the knowledge graph 222 based on past connection coefficients with reference to the reference connection coefficient 223. In Figure 11, white blocks represent nodes and solid lines represent edges. The root (leftmost) node represents a node in the base layer, the terminal (rightmost) node represents a node in the action layer, and the intermediate layer node represents a node in the event layer. The thickness of the solid line represents the strength of the edge connection coefficient, with thicker lines indicating higher strength and thinner lines indicating lower strength.

[0086] As shown in Figure 11(a), the knowledge graph 222 has a first topic T101 and a second topic T102. Here, when the inference device 200 identifies node N101 in the event layer as an event of interest to the user, it can present node N102 in the action layer to the user as a candidate action based on the strength of the connection coefficient.

[0087] According to Figure 11(b), the knowledge graph 222 has a first topic T101 and a second topic T102. For the first topic T101, the connection coefficients included in the first reference connection coefficient 223A are reflected in the edges between nodes. For the second topic T102, the connection coefficients included in the second reference connection coefficient 223B are reflected in the edges between nodes. The first reference connection coefficient 223A is, for example, a topic related to "investment" and has the time attribute of "30s". The second reference connection coefficient 223B is, for example, a topic related to "health" and has the time attribute of "50s". In other words, reference connection coefficients 223 with different time attributes are applied to the knowledge graph 222 depending on the topic. Note that for the second topic T102, node N104 in the behavior layer has disappeared. This means that at the time of the second reference connection coefficient 223B (50s), the behavior candidate corresponding to node N104 was not selected by the user.

[0088] In Figure 11(b), when the inference device 200 identifies node N101 in the event layer as an event of interest to the user, it can present node N103 in the action layer to the user as an action candidate based on the strength of the connection coefficient included in the first reference connection coefficient 223A. Furthermore, if the user attempts to perform an action different from the action candidate of node N103 at the current time, the inference device 200 can output a control signal to prevent that action. In this way, the technology of this disclosure makes it possible to support the current user's decision-making with information on their past accumulated decision-making.

[0089] The disclosures herein include the following decision support systems, apparatus, methods, and computer programs.

[0090] (Item 1) A decision support system that assists a user in making decisions based on a personalized knowledge graph, comprising: a storage unit for storing a hierarchical knowledge graph having a plurality of prioritized base layers; an interest information acquisition unit for acquiring interest information including the user's interest events; an identification unit for identifying a path to a node in the knowledge graph corresponding to the interest events based on the interest information; a proposal unit for determining and presenting candidate actions included in the path to the user based on connection coefficients between the nodes; an action acquisition unit for acquiring a selected action chosen by the user from the candidate actions; an update unit for amplifying the connection coefficients between nodes included in the path to the node corresponding to the selected action in the knowledge graph; and an output unit for outputting the amplified connection coefficients between nodes included in the path to the node corresponding to the selected action to the user. (Item 2) The decision support system according to item 1, wherein the update unit attenuates the connection coefficients between nodes included in the path to the node corresponding to the action candidate that was not selected as the chosen action in the knowledge graph. (Item 3) The decision support system according to item 1 or item 2, wherein the output unit highlights the paths in the knowledge graph on the graphical user interface as the connection coefficient between the nodes increases, and de-highlights the paths in the knowledge graph on the graphical user interface as the connection coefficient between the nodes decreases. (Item 4) The output unit presents to the user, through visual changes in a graphical user interface, the amplification of connection coefficients between nodes included in the path leading to the node corresponding to the selection action, as a decision support system according to any one of items 1 to 3. (Item 5) A decision support system according to any one of items 1 to 4, wherein the proposal unit generates action candidates not included in the knowledge graph based on past connection coefficient trends in the knowledge graph and presents them to the user; the action acquisition unit acquires the selected action chosen by the user for the action candidates not included in the knowledge graph; and the update unit generates a new node corresponding to the action candidate included in the selected action, adds it to the knowledge graph, and amplifies the connection coefficients between nodes included in the path to the node. (Item 6) The decision support system according to any one of items 1 to 5, wherein the interest information acquisition unit acquires the interest information based on the user's gaze information detected by the wearable device, and the output unit presents the knowledge graph as a virtual object superimposed on real space via the transparent display of the wearable device. (Item 7) The decision support system according to any one of items 1 to 6, wherein the update unit stores the update history of the connection coefficients due to the selection behavior in the memory unit, links it to the knowledge graph, evaluates the consistency of the update history of the connection coefficients associated with the selection behavior of a predetermined topic over a predetermined period, and stores the connection coefficients between nodes included in a consistent topic in the memory unit as reference connection coefficients, linked to time attributes. (Item 8) The decision support system according to item 7, comprising: a reference information acquisition unit that acquires reference information from the user, including time attributes of a knowledge graph to be referenced in decision-making, and topics; and a decision-making unit that identifies the time attributes of the reference information having topics corresponding to the interest information acquired from the user, and determines the action candidates based on the standard connection coefficient linked to the time attributes. (Item 9) The decision support system according to item 8, further comprising a control signal output unit that outputs a control signal to inhibit the user's chosen action when the action candidates presented by the decision-making unit conflict with the user's chosen action. (Item 10) A decision support system according to any one of items 7 to 9, comprising a mediation unit in which a mediation agent determines the priority of the reference connection coefficients in dialogue with the user when the reference connection coefficients for a topic corresponding to interest information obtained from the user are associated with multiple time attributes. (Item 11) A decision support method that assists a user in making decisions based on a personalized knowledge graph, wherein a computer performs the following steps: a storage step of storing a hierarchical knowledge graph having a plurality of prioritized base layers; an interest information acquisition step of acquiring interest information including the user's interest events; an identification step of identifying a path to a node in the knowledge graph corresponding to the interest events based on the interest information; a proposal step of determining and presenting candidate actions included in the path to the user based on connection coefficients between the nodes; an action acquisition step of acquiring a selected action chosen by the user from the candidate actions; an update step of amplifying the connection coefficients between nodes included in the path to the node corresponding to the selected action in the knowledge graph; and an output step of outputting the amplified connection coefficients between nodes included in the path to the node corresponding to the selected action to the user. (Item 12) A decision support program that assists a user in making decisions based on a personalized knowledge graph, wherein the computer functions as: a storage unit that stores a hierarchical knowledge graph having a plurality of prioritized base layers; an interest information acquisition unit that acquires interest information including the user's interest events; an identification unit that identifies a path to a node in the knowledge graph corresponding to the interest events based on the interest information; a proposal unit that determines and presents candidate actions included in the path to the user based on connection coefficients between the nodes; an action acquisition unit that acquires the selected action chosen by the user from the candidate actions; an update unit that amplifies the connection coefficients between nodes included in the path to the node corresponding to the selected action in the knowledge graph; and an output unit that outputs the amplified connection coefficients between nodes included in the path to the node corresponding to the selected action to the user.

[0091] This disclosure is not limited to the embodiments described above, and various modifications, improvements, and variations are possible without departing from the spirit and scope of this disclosure. [Explanation of Symbols]

[0092] 1. Decision support system 100 generator 110 processors 111 Interest Information Acquisition Department 112 Specific section 113 Proposal Department 114 Behavior Acquisition Department 115 Update Department 116 Knowledge Analysis Department 117 Output section 120 memory (storage unit) 121 Knowledge Graph 122 Reference connection coefficient 130 Communication Interfaces 200 Reasoning device 210 processors 211 Reference information acquisition unit 212 Interest Information Acquisition Department 213 Decision Making Department 214 Mediation Department 215 Control signal output section 220 memory (storage unit) 221 Reference Information 222 Knowledge Graph 223 Reference connection coefficient 300 user terminals 400 wearable devices NW (Network Communication Network)

Claims

1. A decision support system that assists users in making decisions based on a personalized knowledge graph, A memory unit for storing a hierarchical knowledge graph comprising a knowledge graph consisting of multiple nodes and edges connecting the nodes, the knowledge graph having a plurality of prioritized base layers, an event layer consisting of one or more layers under the base layers, and one or more action layers under the event layer. A unit for acquiring interest information that acquires interest information, including the user's interests, based on data including text, images, or audio input through the user terminal's input interface, or the user's gaze information detected by a wearable device. A selection unit that identifies a path from the node corresponding to the base layer in the knowledge graph to the node corresponding to the event of interest in the event layer below it, based on the aforementioned interest information, A proposal unit that, with respect to the nodes of the action layer connected to the nodes of the event layer included in the aforementioned path, determines one or more action layer nodes as action candidates whose connection coefficient, representing the strength of the connection between these nodes, is equal to or greater than a predetermined threshold, and presents them to the user, An action acquisition unit that acquires the selected action chosen by the user from the aforementioned action candidates, An update unit that amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the selection action in the knowledge graph, A decision support system comprising: an output unit that outputs to the user an amplified connection coefficient between nodes included in the path leading to the node corresponding to the aforementioned selection action.

2. The decision support system according to claim 1, wherein the update unit attenuates the connection coefficients between nodes included in the path to the node corresponding to the action candidate that was not selected as the chosen action in the knowledge graph.

3. The decision support system according to claim 1 or claim 2, wherein the output unit highlights the paths in the knowledge graph on the graphical user interface as the connection coefficient between the nodes increases, and does not highlight the paths in the knowledge graph on the graphical user interface as the connection coefficient between the nodes decreases.

4. The decision support system according to claim 1 or 2, wherein the output unit presents to the user, by visual changes in the graphical user interface, the amplification of connection coefficients between nodes included in the path leading to the node corresponding to the selection action.

5. The proposal unit, based on the trend of past connection coefficients in the knowledge graph, refers to nodes of similar events that are similar to the events of the event layer nodes connected to the node of the action candidate, generates a new node of action candidate connected to the event layer node based on the action corresponding to the node of the action layer connected to the node of the similar event, and presents it to the user. The action acquisition unit acquires the selected action chosen by the user for action candidates not included in the knowledge graph. The decision support system according to claim 1 or 2, wherein the update unit generates a new node corresponding to the action candidate included in the selected action, adds it to the knowledge graph, and amplifies the connection coefficient between nodes included in the path to the node.

6. The interest information acquisition unit acquires the interest information based on the user's gaze information detected by the wearable device. The decision support system according to claim 1 or 2, wherein the output unit presents the knowledge graph as a virtual object superimposed on real space via the transparent display of the wearable device.

7. The update unit stores the update history of the connection coefficients resulting from the selection behavior in the memory unit, linked to the knowledge graph. The system includes a knowledge analysis unit that evaluates the consistency of the update history of the connection coefficients associated with the selection behavior of a predetermined topic over a predetermined period, and stores the connection coefficients between nodes included in a consistent topic as reference connection coefficients in the storage unit, linked to time attributes. The aforementioned topic is an identification metric for defining a partial network of nodes and edges that constitute the knowledge graph, which share a common context or purpose. The aforementioned time attribute is a label representing the base period of the update history subject to the consistency evaluation, The decision support system according to claim 1 or 2, wherein the knowledge analysis unit evaluates the consistency of the update history of the connection coefficients according to at least one consistency index selected from the variance of the connection coefficients, the selection frequency of a particular node, and the reaction time of the selection behavior.

8. A reference information acquisition unit that obtains reference information from the user, including the time attribute of the knowledge graph to be referenced in decision-making, and the topic. The decision support system according to claim 7, comprising: a decision-making unit that identifies the time attribute of the reference information having a topic corresponding to the interest information obtained from the user, and determines the candidate action based on the reference connection coefficient associated with the time attribute.

9. The decision support system according to claim 8, further comprising a control signal output unit that outputs a control signal to inhibit the user's chosen action when the action candidate presented by the decision-making unit conflicts with the user's chosen action.

10. The decision support system according to claim 7, further comprising a mediation unit in which a mediation agent determines the priority of the reference connection coefficients in dialogue with the user when the reference connection coefficients for a topic corresponding to interest information obtained from the user are associated with multiple time attributes.

11. A decision support method that assists users in making decisions based on a personalized knowledge graph, Computers A memory process for storing a hierarchical knowledge graph comprising a knowledge graph consisting of multiple nodes and edges connecting the nodes, the knowledge graph having a plurality of prioritized base layers, an event layer consisting of one or more layers under the base layers, and one or more action layers under the event layer. A step of acquiring interest information, which includes data including text, images, or audio input through the input interface of a user terminal, or the user's gaze information detected by a wearable device, and which acquires interest information including the user's interests. A process of identifying a path from the node corresponding to the base layer in the knowledge graph to the node corresponding to the event of interest in the event layer below it, based on the aforementioned interest information, A proposal step of selecting one or more action layer nodes as action candidates, where the connection coefficient representing the strength of the connection between these nodes is equal to or greater than a predetermined threshold, with respect to the action layer nodes connected to the event layer nodes included in the aforementioned path, and presenting these to the user, An action acquisition step to acquire the selected action chosen by the user from the aforementioned action candidates, An update process that amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the selection action in the knowledge graph, A decision support method that performs an output step of outputting to the user an amplification of the connection coefficients between nodes included in the path leading to the node corresponding to the selection action.

12. A decision support program that assists users in making decisions based on a personalized knowledge graph, Computers, A memory unit for storing a hierarchical knowledge graph comprising a knowledge graph consisting of multiple nodes and edges connecting the nodes, the knowledge graph having a plurality of prioritized base layers, an event layer consisting of one or more layers under the base layers, and one or more action layers under the event layer. A unit for acquiring interest information that acquires interest information, including the user's interests, based on data including text, images, or audio input through the user terminal's input interface, or the user's gaze information detected by a wearable device. A selection unit that identifies a path from the node corresponding to the base layer in the knowledge graph to the node corresponding to the event of interest in the event layer below it, based on the aforementioned interest information, A proposal unit that, with respect to the nodes of the action layer connected to the nodes of the event layer included in the aforementioned path, determines one or more action layer nodes as action candidates whose connection coefficient, representing the strength of the connection between these nodes, is equal to or greater than a predetermined threshold, and presents them to the user, An action acquisition unit that acquires the selected action chosen by the user from the aforementioned action candidates, An update unit that amplifies the connection coefficients between nodes included in the path leading to the node corresponding to the selection action in the knowledge graph, A decision support program that functions as an output unit that outputs to the user the amplification of connection coefficients between nodes included in the path leading to the node corresponding to the aforementioned selection action.

Images