User interaction method and device based on intelligent agent, equipment, medium and product

Abstract

The invention provides an agent-based user interaction method and device, equipment, a medium and a product, which can be applied to the technical field of artificial intelligence and the field of financial science and technology. The method comprises the steps of updating a scene sub-graph in a preset knowledge graph according to current input information of a target user for a preset agent; the scene sub-graph is used for storing historical interaction information of the target user and the preset intelligent agent; according to the semantic entities in the updated scene sub-graph and the incidence relation between the different semantic entities, updating a semantic entity sub-graph in a preset knowledge graph; performing semantic entity clustering on the updated semantic entity sub-graph, and updating a community sub-graph in a preset knowledge graph according to a semantic entity clustering result; searching a target node related to the current input information in a preset knowledge graph; determining memory prompt information according to the retrieved target node; and for the current input information and the memory prompt information, generating feedback information based on a preset agent.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence technology and fintech, specifically to a user interaction method, device, equipment, medium, and product based on intelligent agents. Background Technology

[0002] With the widespread adoption of smart devices, users can interact with them in various ways. For example, users can ask questions to applications or smart agents installed on their smart devices and receive answers.

[0003] However, current user interactions typically only process information from a single interaction, resulting in low accuracy and a poor user experience. Summary of the Invention

[0004] In view of the above problems, this application provides a user interaction method, apparatus, device, medium and product based on intelligent agents to improve the accuracy of user interaction and enhance the user interaction experience.

[0005] According to a first aspect of this application, a user interaction method based on an intelligent agent is provided, comprising: updating a scenario subgraph in a current preset knowledge graph based on current input information from a target user to a preset intelligent agent; the scenario subgraph being used to store historical interaction information between the target user and the preset intelligent agent; updating a semantic entity subgraph in the current preset knowledge graph based on semantic entities in the updated scenario subgraph and the relationships between different semantic entities; any node in the semantic entity subgraph representing a semantic entity; any side in the semantic entity subgraph representing the relationships between semantic entities represented by different connected nodes; performing semantic entity clustering on the updated semantic entity subgraph, and updating a community subgraph in the current preset knowledge graph based on the semantic entity clustering results; any node in the community subgraph representing a semantic entity cluster, and any side in the community subgraph representing the relationships between semantic entity clusters represented by different connected nodes; retrieving a target node related to the current input information in the current preset knowledge graph; determining memory prompt information based on the retrieved target node; and generating feedback information for the current input information based on the preset intelligent agent, in response to the current input information and the memory prompt information.

[0006] Optionally, the scenario subgraph is also used to store interaction context information of different historical interaction information; the step of updating the scenario subgraph in the current preset knowledge graph according to the target user's current input information to the preset agent includes: updating the scenario subgraph in the current preset knowledge graph according to the target user's current input information to the preset agent and the interaction context information of the current input information.

[0007] Optionally, the method further includes: identifying semantic entities and the relationships between different semantic entities for the updated scenario subgraph; performing contradiction detection on the identified relationships; determining the chronological order of the contradictory relationships when it is determined that contradictory relationships are identified for any two semantic entities; and updating the semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities, including: updating the semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph, the relationships between different semantic entities, and the chronological order of the determined relationships.

[0008] Optionally, retrieving target nodes related to the current input information in the current preset knowledge graph includes: retrieving candidate nodes related to the current input information in the current preset knowledge graph using different retrieval methods to obtain different candidate node sets; and determining the target node related to the current input information based on the union of the obtained different candidate node sets.

[0009] Optionally, retrieving target nodes related to the current input information in the current preset knowledge graph includes at least one of the following: retrieving target nodes related to the current input information in the current preset knowledge graph based on the similarity between the current input information and semantic entity information; retrieving target nodes related to the current input information in the current preset knowledge graph based on the similarity between the current input information and semantic entity cluster summary information; wherein the semantic entity cluster summary information is determined based on the information of semantic entities in the corresponding semantic entity cluster; and retrieving target nodes related to the current input information in the current preset knowledge graph based on nodes used to represent any semantic entity in the current input information.

[0010] Optionally, determining the memory prompt information based on the retrieved target nodes includes: among the retrieved target nodes, determining the memory prompt information based on the target nodes whose correlation with the current input information is greater than a preset correlation threshold.

[0011] Optionally, the method for determining the degree of association between the target node and the current input information includes at least one of the following: determining the degree of association between the target node and the current input information based on the frequency of occurrence of the target node in different search results; the target node being determined based on different search results obtained from different search methods; determining the degree of association between the target node and the current input information based on the node distance between the target node and the node used to represent the current input information; and determining the degree of association between the target node and the current input information based on a preset large language model.

[0012] A second aspect of this application provides a user interaction device based on an intelligent agent, comprising: a scenario subgraph module, used to update a scenario subgraph in a current preset knowledge graph based on the current input information of a target user to a preset intelligent agent; the scenario subgraph is used to store historical interaction information between the target user and the preset intelligent agent; and a semantic entity module, used to update a semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities; any node in the semantic entity subgraph is used to represent a semantic entity; any side in the semantic entity subgraph is used to represent the relationship between the semantic entities represented by the different connected nodes. The system includes a community subgraph module for performing semantic entity clustering on the updated semantic entity subgraph and updating the community subgraph in the current preset knowledge graph based on the semantic entity clustering results. Each node in the community subgraph represents a semantic entity cluster, and each edge in the community subgraph represents the association between semantic entity clusters represented by different connected nodes. A processing module is used to retrieve target nodes related to the current input information in the current preset knowledge graph; determine memory prompt information based on the retrieved target nodes; and generate feedback information for the current input information based on the preset agent, taking into account the current input information and the memory prompt information.

[0013] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0016] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0017] Figure 1 This illustration schematically depicts an application scenario of the agent-based user interaction method according to an embodiment of this application.

[0018] Figure 2 A flowchart illustrating an agent-based user interaction method according to an embodiment of this application is shown schematically.

[0019] Figure 3 This schematically illustrates a structural block diagram of an agent-based user interaction device according to an embodiment of the present application;

[0020] Figure 4 A block diagram schematically illustrates an electronic device suitable for implementing an agent-based user interaction method according to an embodiment of this application. Detailed Implementation

[0021] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0022] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0023] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0024] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0025] With the widespread adoption of smart devices, users can interact with them in various ways. For example, users can ask questions to apps or smart agents installed on their devices and receive answers. However, current user interactions typically only process information from a single interaction, resulting in a poor user experience. For instance, when a user inputs "explain the previously mentioned concept in detail," feedback often fails to consider historical interactions, leading to ineffective responses and a poor user experience.

[0026] To address the aforementioned technical problems, embodiments of this application provide a user interaction method based on an intelligent agent.

[0027] This method can improve the interaction process between the user and the agent. Specifically, it can combine the current interaction information or the user's current input information with the historical interaction content between the user and the agent, extract historical content related to the current interaction, and use it as prompt information for the agent to generate feedback information. This can improve the accuracy and comprehensiveness of the interaction content between the agent and the user, thereby improving the accuracy of the user interaction and enhancing the user experience.

[0028] For example, in response to the user's current input of "explain the previously mentioned concept A in detail", relevant content about concept A can be retrieved from historical interaction content. This content can then be used as a prompt to help the agent generate more accurate feedback, thereby improving the agent's memory capabilities and enhancing the user's interactive experience.

[0029] In this method, historical interaction content can be organized, specifically by determining a knowledge graph based on the historical interaction content, which allows for more convenient and faster retrieval within the knowledge graph, improving retrieval efficiency and accuracy.

[0030] It's understandable that the current interaction might contain some factual or supporting information. For example, a user might input a document that needs to be analyzed. The content of the document might be related to or conflict with historical interaction content. Therefore, updating the knowledge graph based on the information from the current interaction allows for real-time updates, improving the timeliness and accuracy of knowledge within the knowledge graph. This, in turn, enhances retrieval accuracy and improves the user experience.

[0031] It should be noted that the agent-based user interaction method and apparatus provided in the embodiments of this application can be applied to the fields of artificial intelligence and fintech, and can also relate to the application of large models in human-computer interaction scenarios. For example, the agent-based user interaction method provided in the embodiments of this application can be used to achieve user interaction for agents in financial institutions or banks. The embodiments of this application can also be applied to any field other than fintech, such as agent user interaction in the question-and-answer or education fields. The application fields of the agent-based user interaction method and apparatus provided in the embodiments of this application are not limited.

[0032] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0033] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0034] Figure 1 The diagram illustrates an application scenario of the agent-based user interaction method according to an embodiment of this application.

[0035] like Figure 1 As shown, application scenario 100 according to this embodiment may include: a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0036] Users can use the first terminal device 101, the second terminal device 102, or the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0038] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, or the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0039] It should be noted that the agent-based user interaction method provided in this application embodiment can generally be executed by server 105. Correspondingly, the agent-based user interaction device provided in this application embodiment can generally be located in server 105. The agent-based user interaction method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the agent-based user interaction device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0040] It is understood that server 105 can execute the agent-based user interaction method provided in this application embodiment in response to the interactive information sent by the user through the terminal device, and feed back the generated feedback information to the terminal device. Agents or interfaces for communicating with agents can be deployed in server 105. The agent-based user interaction method provided in this application embodiment can also be executed by the terminal device. Agents or interfaces for communicating with agents can be deployed in the terminal device.

[0041] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0042] Figure 2 A flowchart illustrating an agent-based user interaction method according to an embodiment of this application is shown.

[0043] like Figure 2As shown, this embodiment provides a user interaction method based on an intelligent agent, which may include operations S210 to S240. This method flow is not limited to a specific executing entity; optionally, it can be applied to any device or any application. For example, it can be applied to an intelligent agent or a preset intelligent agent, or it can be applied to a server or terminal, etc.

[0044] In operation S210, the scenario subgraph in the current preset knowledge graph is updated based on the current input information of the target user to the preset intelligent agent; the scenario subgraph is used to store the historical interaction information between the target user and the preset intelligent agent.

[0045] In operation S220, based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities, the semantic entity subgraph in the current preset knowledge graph is updated; any node in the semantic entity subgraph is used to represent a semantic entity; any side in the semantic entity subgraph is used to represent the relationships between the semantic entities represented by the different nodes connected to it.

[0046] In operation S230, semantic entity clustering is performed on the updated semantic entity subgraph, and the community subgraph in the current preset knowledge graph is updated according to the semantic entity clustering results. Any node in the community subgraph is used to represent a semantic entity cluster, and any side in the community subgraph is used to represent the association between the semantic entity clusters represented by the different connected nodes.

[0047] In operation S240, target nodes related to the current input information are retrieved from the current preset knowledge graph; memory prompt information is determined based on the retrieved target nodes; and feedback information is generated based on the preset agent for the current input information and memory prompt information.

[0048] This method can combine historical interaction information from a pre-set knowledge graph to determine memory prompts. By combining the current input information and memory prompts with a pre-set intelligent agent to generate feedback information, the accuracy and comprehensiveness of the interaction between the pre-set intelligent agent and the user can be improved, thereby enhancing the accuracy of user interaction and the user experience.

[0049] This method can also improve retrieval efficiency and accuracy by searching a pre-defined knowledge graph containing historical interaction information.

[0050] This method can also update the preset knowledge graph containing historical interaction information based on the current input information. This allows for the easy integration of knowledge or information from the current input information into the preset knowledge graph, improving the real-time performance and accuracy of the knowledge in the preset knowledge graph. This, in turn, can be combined with the retrieval steps to improve the accuracy of the retrieval, further enhance the accuracy of memory prompts and feedback information, and improve the user interaction experience.

[0051] In the embodiments of this application, user consent or authorization can be obtained for user information, such as information entered by the user. If user consent or authorization allows the acquisition of user information, then user information can be acquired.

[0052] In the embodiments of this application, a corresponding operation entry point can be provided to the user, allowing the user to choose to agree to or reject the automated decision result. That is, before processing user information (such as currently input information) to realize user interaction, the user's instruction to agree or refuse to use user information for user interaction can be obtained through the corresponding operation entry point. If the user agrees to use user information for user interaction, then user interaction can be carried out using user information. If the user refuses to use user information for user interaction, then the expert decision-making process is initiated.

[0053] The embodiments of this application are not limited to a target user. Optionally, the target user can be any user; for ease of description, any user interacting with the preset intelligent agent will be referred to as the target user.

[0054] The embodiments of this application are not limited to a pre-defined intelligent agent. Optionally, the pre-defined intelligent agent can be any intelligent agent, which can be an artificial intelligence system capable of perceiving the environment, making autonomous decisions, and executing actions to achieve a specific goal. Specifically, it can be a pre-defined large language model, or an application or interface used to implement functions using the large language model. It is understood that many operations and steps in the embodiments of this application can be implemented using a large language model.

[0055] The embodiments of this application are not limited to a pre-defined knowledge graph. Optionally, the pre-defined knowledge graph may include a context subgraph, a semantic entity subgraph, and a community subgraph. These three subgraphs can be used to summarize interaction information, facilitating retrieval of interaction information. It is understood that the pre-defined knowledge graph can be updated in real time, specifically by adding more nodes and relationships as the interaction between the target user and the pre-defined intelligent agent gradually increases.

[0056] The embodiments of this application are not limited to historical interaction information. Optionally, the historical interaction information may include the target user's historical input information to the preset intelligent agent, and may also include the preset intelligent agent's historical feedback information. This allows for convenient combination of the interaction information between the target user and the preset intelligent agent, improving the accuracy of the feedback information generated by the preset intelligent agent and thus improving the accuracy of user interaction.

[0057] It is understood that the above method embodiments update the preset knowledge graph based on the target user's current input information to the preset intelligent agent. The preset knowledge graph may contain historical interaction information between the target user and the preset intelligent agent, and relevant information is retrieved from the updated preset knowledge graph as memory prompts. This allows the preset intelligent agent to combine the current input information and the memory prompts, thereby generating feedback information using relevant information from the historical interaction information. This can improve the preset intelligent agent's memory ability, improve the accuracy and comprehensiveness of the generated feedback information, and enhance the user interaction experience.

[0058] For example, specific memory prompts may include the abbreviation of a concept in the current input information. The full name of the concept can be determined by searching historical interaction information, and thus the full name of the concept can be included in the memory prompts, making it easier for the preset agent to generate more accurate feedback information.

[0059] The following explanation addresses several aspects of the embodiments of this application.

[0060] 1. In operation S210, the scenario subgraph in the current preset knowledge graph is updated according to the current input information of the target user to the preset intelligent agent; the scenario subgraph is used to store the historical interaction information between the target user and the preset intelligent agent.

[0061] The embodiments of this application are not limited to the current input information. Optionally, the current input information can be in text, voice, or multimedia form. For example, the target user can input documents, audio, video, or other information to a preset intelligent agent.

[0062] The embodiments of this application are not limited to scenario subgraphs. Scenario subgraphs can be used to store historical interaction information between a target user and a preset intelligent agent, specifically, they can be used to store the original information of the historical interaction information between the target user and the preset intelligent agent. For example, the original version of the target user's historical input information to the preset intelligent agent, as well as the original information of the preset intelligent agent's feedback to the target user, can be stored as historical interaction information in the scenario subgraph, which can conveniently preserve the original interaction information between the target user and the preset intelligent agent.

[0063] The embodiments of this application do not limit the specific content stored in the scenario subgraph. Optionally, the scenario subgraph can also be used to store context information of historical interaction information, which can facilitate the preservation of more comprehensive historical interaction information. The embodiments of this application do not limit the context information of the interaction information. Optionally, specific context information may include, for example, business operation information before or after the target user interacts with the preset intelligent agent, or operation information of the target user during the interaction with the preset intelligent agent.

[0064] In a specific example, the target user can approve or like the information provided by the preset intelligent agent, which can be used as business operation information or context information. Before interacting with the preset intelligent agent, the target user may be executing a certain business process, which can determine that the target user may have questions about the business process, and thus the information of the business process can be identified as context information.

[0065] Therefore, optionally, the scenario subgraph can also be used to store interaction context information of different historical interactions. Updating the scenario subgraph in the current preset knowledge graph based on the target user's current input information to the preset agent can specifically include: updating the scenario subgraph in the current preset knowledge graph based on the target user's current input information to the preset agent and the interaction context information of the current input information. This embodiment can update the preset knowledge graph by combining interaction context information, which can improve the memory capacity of the preset agent, improve the comprehensiveness and accuracy of retrieval, increase the amount and scope of information for determining memory prompts, and improve the comprehensiveness and accuracy of memory prompts.

[0066] The embodiments of this application do not limit the interaction context information. Optionally, the interaction context information may include the target user's business information or business operation information. Specifically, it may include the business information or business operation information involved by the target user during, before, or after interacting with the preset intelligent agent.

[0067] The embodiments of this application do not limit the specific method of updating the scenario subgraph based on the current input information. Optionally, the scenario subgraph can be constructed based on the current input information and historical interaction information; alternatively, it can be updated based on the current scenario subgraph and the current input information. Specifically, new nodes can be generated based on the current input information, added to the scenario subgraph, and edges can be generated between the new nodes and existing nodes in the scenario subgraph. Specifically, edges can be constructed between the new nodes and nodes representing the previous interaction information. Correspondingly, the embodiments of this application do not limit the specific method of updating the scenario subgraph based on the current input information and the interaction context information of the current input information. Optionally, new nodes can be generated based on the current input information and the interaction context information of the current input information, added to the scenario subgraph, and edges connecting the new stage to existing nodes in the scenario subgraph can be generated. The generated new nodes can contain the current input information and the interaction context information of the current input information, or two new nodes can be generated, representing the current input information and the interaction context information of the current input information, respectively.

[0068] The embodiments of this application do not limit the specific structure of the scenario subgraph. Optionally, any node in the scenario subgraph can be used to represent single historical interaction information, and any edge in the scenario subgraph can be used to represent the temporal order of the interaction information. The scenario subgraph may contain nodes representing single interaction information, and may also contain nodes representing interaction context information. Accordingly, the edges between different nodes representing single interaction information can represent the temporal relationship between the interaction information; the edges between a node representing interaction context information and a node representing single interaction information can be used to represent that the interaction context information is a context for that interaction information.

[0069] In one optional embodiment, the scenario subgraph can be updated for each interaction between the target user and the preset intelligent agent, and the information of each interaction can be stored in the scenario subgraph, so that the scenario subgraph can store the historical interaction information between the target user and the preset intelligent agent.

[0070] Optionally, the current input information can be stored as new information in the context subgraph, and this step can be repeated. When the current input information is detected, the context subgraph can be updated according to the current input information; or, in response to receiving the current input information from the target user for the preset agent, the context subgraph can be updated according to the current input information.

[0071] Second, in operation S220, based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities, update the semantic entity subgraph in the current preset knowledge graph; any node in the semantic entity subgraph is used to represent a semantic entity; any side in the semantic entity subgraph is used to represent the relationships between the semantic entities represented by the different nodes connected to it.

[0072] The embodiments of this application do not limit semantic entities, nor do they limit the relationships between semantic entities. Optionally, a semantic entity can be a linguistic unit that represents concrete or abstract things in the real world and has clear semantic information, such as named entities, abstract concepts, geographical entities, event names, etc. Optionally, the relationships between semantic entities can specifically represent the relationships between concrete or abstract things; for example, a semantic entity can belong to another semantic entity, or a semantic entity can be an antonym of another semantic entity, etc.

[0073] It is understandable that semantic entities, and the relationships between semantic entities, can be extracted from the context subgraph. Specifically, they can be extracted from the interaction information contained in the context subgraph, and can reflect some facts or information relationships in the interaction information.

[0074] The embodiments of this application do not limit the specific process of updating the semantic entity subgraph.

[0075] Optionally, semantic entities can be identified within the updated context subgraph, and the relationships between different semantic entities can be identified. The semantic entity subgraph can then be updated based on the identified semantic entities and the relationships between them.

[0076] Optionally, a semantic entity subgraph can be constructed based on the updated context subgraph; alternatively, the current semantic entity subgraph can be updated based on the updated context subgraph. Specifically, the current semantic entity subgraph can be updated based on the updated content in the updated context subgraph, which may involve adding new nodes and edges, or adjusting the content of nodes or edges, etc.

[0077] In one optional embodiment, the semantic entity subgraph can be further updated for each updated context subgraph, thereby updating the context subgraph accordingly for multiple interactions between the target user and the preset agent, storing the information of each interaction in the context subgraph, and further updating the semantic entity subgraph accordingly based on each updated context subgraph.

[0078] Understandably, the semantic entity subgraph is updated based on the identified semantic entities and the relationships between different semantic entities. Specifically, this can be done by updating the current semantic entity subgraph. For example, based on the identification results, the semantic entities or relationships in the current semantic entity subgraph that have been updated can be determined and updated; or based on the identification results, semantic entities or relationships that do not exist in the current semantic entity subgraph can be determined and updated, and so on.

[0079] In one optional embodiment, it is considered that there may be contradictory information or information that changes over time in historical interaction information and current input information. For example, for the same pair of semantic entities, the historical interaction information may determine that they are in an inclusive relationship, while the current input information may update it to a disjoint relationship, meaning that there is no intersection between these two semantic entities, thus identifying contradictory information. Therefore, information that changes over time can be marked to facilitate subsequent retrieval and the determination of information that changes over time by the pre-defined intelligent agent.

[0080] Optionally, time information can be labeled for each identified semantic entity and the relationships between different semantic entities. The embodiments of this application do not limit the specific method of determining the corresponding time information. Optionally, it can be determined based on the time information of the extracted information source (historical interaction information or current input information). Alternatively, the time information of the source of the semantic entity or relationship information can be extracted and used as the corresponding time information for labeling. This facilitates the determination of how information changes over time, improving the accuracy of subsequent retrieval and the generation of feedback information by the preset intelligent agent.

[0081] Optionally, the above method may further include: identifying semantic entities and the relationships between different semantic entities for the updated scenario subgraph; performing contradiction detection on the identified relationships; and determining the chronological order of the contradictory relationships when different contradictory relationships are identified for any two semantic entities. Accordingly, the semantic entity subgraph in the current preset knowledge graph is updated based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities. Specifically, this may include updating the semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph, the relationships between different semantic entities, and the determined chronological order of the different relationships. This embodiment, through contradiction detection and determining the chronological order of contradictory relationships, can easily clarify how relationships change over time, improve the accuracy of the extracted semantic information, facilitate improved accuracy of subsequent retrieval, and improve the accuracy of feedback information generated by the preset intelligent agent, while reducing the impact of contradictory information.

[0082] The embodiments of this application do not limit the specific method of contradiction detection. Optionally, when multiple different relationships are identified between any two semantic entities, the semantic information of these multiple different relationships can be analyzed to determine whether a contradiction exists. For example, the semantic information of multiple different relationships may contain contradictory semantic information such as "belongs to" and "does not belong to". Specifically, different concepts may have the same abbreviation; concept A and concept B may both be abbreviated as C. In historical interaction information, it is believed that the abbreviation C belongs to a certain concept A. However, as interactions increase, the abbreviation C changes to the abbreviation of another concept B, belonging to concept B but not concept A. Thus, it can be determined that there are two contradictory relationships of "belonging to" and "not belonging to" between the semantic entity with the abbreviation and the semantic entity with concept A. Specifically, this can be done by analyzing multiple different relationships identified between any two semantic entities based on a preset large language model to determine whether a contradiction exists.

[0083] Understandably, when dealing with contradictory relationships, the relationship that comes first in chronological order can be identified as an older or changed fact, while the most recent relationship in chronological order can be identified as the current fact. Therefore, different relationships can be labeled according to chronological order to facilitate the identification of the current fact. In subsequent searches, information can be expanded by incorporating older facts, or only the current fact can be considered to improve the accuracy and real-time performance of the search. During the process of generating feedback information by the pre-set intelligent agent, the amount of information generated in the feedback information can be expanded by incorporating changes in older facts or relationships to improve comprehensiveness, or the focus can be placed on the current fact to improve accuracy; the choice can be made based on actual needs.

[0084] Optionally, the chronological order or corresponding time information can be determined for any one or more identified relationships to facilitate the determination of the effective time of the relationships. The specific method for determining the chronological order or time information is not limited; it can be determined based on the time information of the information source (interaction information). Optionally, corresponding time information can also be determined for any one or more identified semantic entities.

[0085] Third, in operation S230, semantic entity clustering is performed on the updated semantic entity subgraph, and the community subgraph in the current preset knowledge graph is updated according to the semantic entity clustering results; any node in the community subgraph is used to represent a semantic entity cluster, and any side in the community subgraph is used to represent the association relationship between the semantic entity clusters represented by the different connected nodes.

[0086] The embodiments of this application are not limited to a specific method for clustering semantic entities. Optionally, clustering can be performed based on the information or features of semantic entities, or it can be performed by combining the relationships between semantic entities.

[0087] The embodiments of this application do not limit the specific way of updating the community subgraph.

[0088] Optionally, a community subgraph can be constructed based on the semantic entity clustering results for the updated semantic entity subgraph. Specifically, any semantic entity cluster in the semantic entity clustering results can be identified as a community node, and a community subgraph can be constructed based on the community nodes. Alternatively, the current community subgraph can be updated based on the semantic entity clustering results for the updated semantic entity subgraph. Specifically, the semantic entity nodes contained in the community nodes can be updated, or the edges in the current community subgraph can be updated, etc.

[0089] In one optional embodiment, the community subgraph can be further updated for each update of the semantic entity subgraph, thereby updating the context subgraph for multiple interactions between the target user and the preset agent, storing the information of each interaction in the context subgraph, and further updating the semantic entity subgraph according to each updated context subgraph, and then updating the community subgraph according to each updated semantic entity subgraph.

[0090] Understandably, semantic entity clustering, by creating semantic entity clusters, allows similar semantic entities to be grouped together, facilitating the discovery of hidden relationships between them. It also helps to uncover the target user's true or hidden needs based on similar semantic entities they interact with, enabling the pre-defined intelligent agent to generate more accurate feedback. In a specific example, considering the various semantic entities mentioned by the target user—such as mobile phone, computer, and tablet—semantic entity clustering can identify electronic devices or smart devices as a semantic entity cluster. This determines whether the target user is interested in electronic devices or smart devices, allowing the pre-defined intelligent agent to generate feedback based on the information from these semantic entity clusters.

[0091] Therefore, optionally, for nodes in the community subgraph, relevant information about semantic entities in the represented semantic entity cluster can be further summarized to obtain summary information, which facilitates subsequent retrieval and generation of feedback information.

[0092] The embodiments of this application do not limit the edges in the community subgraph, nor do they limit the association relationships between semantic entity clusters. Optionally, edges in the community subgraph can be constructed according to the degree of association between semantic entity clusters. Any edge in the community subgraph is used to represent the degree of association between the semantic entity clusters represented by the different nodes connected. The degree of association can be determined based on the summary information of the semantic entity clusters, or it can be determined based on the association relationship or degree of association between different semantic entities belonging to different semantic entity clusters.

[0093] Fourth, in operation S240, in the current preset knowledge graph, the target node related to the current input information is retrieved; the memory prompt information is determined based on the retrieved target node; and feedback information for the current input information is generated based on the preset intelligent agent for the current input information and the memory prompt information.

[0094] The embodiments of this application are not limited to the specific methods and processes of retrieving target nodes related to the current input information in the current preset knowledge graph.

[0095] Optionally, for at least one of the context subgraph, semantic entity subgraph, and community subgraph of the current preset knowledge graph, target nodes related to the current input information can be retrieved. Different retrieval methods can also be used, such as retrieving historical interaction information similar to the current input information in the context subgraph, retrieving semantic entity nodes similar to or related to the semantic entities in the current input information in the semantic entity subgraph, and retrieving other semantic entity nodes belonging to the same community node as the semantic entities in the current input information in the community subgraph, etc.

[0096] In one optional embodiment, retrieving target nodes related to the current input information within a current preset knowledge graph can specifically include: retrieving candidate nodes related to the current input information using different retrieval methods within the current preset knowledge graph, obtaining different candidate node sets; and determining the target node related to the current input information based on the union of the obtained different candidate node sets. This embodiment can merge the retrieval results of multiple retrieval methods, which can improve the comprehensiveness and accuracy of retrieving information related to the current input information. The embodiments of this application are not limited to the different retrieval methods. For details, please refer to the explanations of other embodiments.

[0097] The embodiments of this application do not limit the specific method of determining the target node based on different candidate node sets. Optionally, the target node related to the current input information can be determined based on the intersection of the obtained different candidate node sets; or the target node related to the current input information can be determined by filtering the intersection or union of different candidate node sets. The specific filtering method is not limited, and can be referred to the explanation of other embodiments. For example, filtering can be performed based on the relevance or association with the current input information.

[0098] Optionally, in the current preset knowledge graph, retrieving target nodes related to the current input information includes at least one of the following: (1) in the current preset knowledge graph, retrieving target nodes related to the current input information based on the similarity between the current input information and semantic entity information; (2) in the current preset knowledge graph, retrieving target nodes related to the current input information based on the similarity between the current input information and semantic entity cluster summary information; the semantic entity cluster summary information is determined based on the information of semantic entities in the corresponding semantic entity cluster; (3) in the current preset knowledge graph, retrieving target nodes related to the current input information based on the nodes used to represent any semantic entity in the current input information. This embodiment can provide multiple retrieval methods, which can improve the comprehensiveness and accuracy of retrieving information related to the current input information. It is understood that the three retrieval method examples given in this embodiment can be combined with the above embodiments to obtain different candidate node sets through multiple retrieval methods.

[0099] The embodiments of this application do not limit the method of determining the similarity between the current input information and the semantic entity information. Optionally, the semantic entity may correspond to semantic entity description information, and the similarity may be determined between the current input information and the semantic entity description information; alternatively, the similarity may be determined between the semantic entity in the current input information and the semantic entity in the current preset knowledge graph. Accordingly, the semantic entity node corresponding to the semantic entity information with a similarity greater than a preset similarity threshold may be determined as the target node.

[0100] The embodiments of this application do not limit the method for determining the similarity between the current input information and the semantic entity cluster summary information, nor do they limit the method for determining the semantic entity cluster summary information. Optionally, the description information of semantic entities in the semantic entity cluster can be summarized using a large language model to obtain the semantic entity cluster summary information, and then the similarity between the current input information and the semantic entity cluster summary information can be determined. In a specific example, determining the similarity between the current input information and the semantic entity cluster summary information can be done by determining the keyword similarity between the current input information and the semantic entity cluster summary information. Specifically, the keyword similarity can be determined based on the keywords in the current input information and the keywords in the semantic entity cluster summary information. Accordingly, community nodes corresponding to semantic entity cluster summary information with a similarity greater than a preset similarity threshold can be determined as target nodes, or semantic entity nodes contained in the semantic entity cluster corresponding to semantic entity cluster summary information with a similarity greater than a preset similarity threshold can be determined as target nodes.

[0101] The embodiments of this application do not limit the specific method and process of retrieving target nodes based on nodes used to represent any semantic entity in the current input information. Optionally, nodes used to represent any semantic entity in the current input information may include nodes in the current preset knowledge graph that have been updated according to the current input information. Specifically, they may include nodes representing the current input information in the scenario subgraph, nodes representing any semantic entity in the current input information in the semantic entity subgraph, and nodes containing any semantic entity in the current input information in the community subgraph. Optionally, other nodes connected to the nodes used to represent any semantic entity in the current input information may be determined as target nodes. Alternatively, target nodes may be retrieved based on nodes used to represent the current input information. For a detailed explanation, please refer to the above embodiments.

[0102] Of course, besides the search method examples above, target nodes can also be retrieved using other search methods. The above search method examples are for illustrative purposes only.

[0103] The embodiments of this application do not limit the specific methods and processes for determining memory prompt information based on the retrieved target nodes.

[0104] Optionally, once a target node is determined, memory prompts can be determined based on the information represented by the target node. Specifically, the information represented by each target node can be used as the memory prompt. Furthermore, filtering can be performed on the target node or the information represented by the target node to improve the accuracy of the memory prompts.

[0105] Therefore, optionally, determining memory prompt information based on the retrieved target nodes can specifically include: among the retrieved target nodes, determining memory prompt information based on target nodes whose correlation with the current input information is greater than a preset correlation threshold. This embodiment can filter target nodes based on their correlation with the current input information, selecting target nodes with a high correlation to determine memory prompt information. This improves the accuracy of memory prompt information, reduces the length and amount of information in subsequent memory prompts for the preset agent, reduces the computational load of the preset agent in generating feedback information, and improves the efficiency, accuracy, and precision of the preset agent in generating feedback information. It is understood that the retrieved target nodes can be those determined based on different candidate node sets in the above embodiment, or they can be determined by filtering the intersection or union of different candidate node sets according to the above filtering method (correlation with the current input information).

[0106] The embodiments of this application do not limit the specific method for determining the degree of association between the target node and the current input information. Optionally, the degree of association or similarity between the information represented by the target node and the current input information can be determined, or the node distance between the target node and the node representing the current input information in the current preset knowledge graph can be determined.

[0107] In an optional embodiment, the method for determining the degree of association between the target node and the current input information includes at least one of the following: (1) determining the degree of association between the target node and the current input information based on the frequency of occurrence of the target node in different search results; the target node is determined based on different search results obtained by different search methods; (2) determining the degree of association between the target node and the current input information based on the node distance between the target node and the node used to represent the current input information; (3) determining the degree of association between the target node and the current input information based on a preset large language model. This embodiment can determine the degree of association between the target node and the current input information in multiple ways, which can improve the accuracy and flexibility of determining the degree of association between the target node and the current input information. Combined with the above embodiment of filtering target nodes, it can be convenient to filter out target nodes with a high degree of association with the current input information to determine the memory prompt information.

[0108] It's understandable that if the same target node frequently appears in the search results of different search methods, it indicates a high degree of relevance between that node and the current input information. Therefore, the frequency of a target node's appearance in different search results can be positively correlated with the degree of relevance between the target node and the current input information.

[0109] The embodiments of this application do not limit the method of determining node distance. Optionally, node distance can be determined based on the number of nodes between nodes, or it can be determined based on the similarity of node representation information. Of course, the degree of association between the target node and the current input information can also be determined directly based on the similarity between the target node representation information and the current input information.

[0110] The embodiments of this application are not limited to nodes used to represent the current input information. Optionally, they can be nodes used to represent part of the information in the current input information. Specifically, they can include nodes that represent any semantic entity in the current input information, or community nodes that contain any semantic entity in the current input information, etc.

[0111] Furthermore, the degree of association between the target node and the current input information can be determined through a pre-set large language model. Specifically, the information represented by the target node and the current input information can be analyzed using the pre-set large language model, and the degree of association between the target node and the current input information can be determined by combining this with a pre-set knowledge graph.

[0112] The embodiments of this application are not limited to the specific process of generating feedback information for the current input information and the memory prompt information based on the preset intelligent agent.

[0113] Optionally, the preset agent can be deployed locally on the executing entity, or the current input information and memory prompt information can be sent to the preset agent through the communication interface of the preset agent, and feedback information generated by the preset agent for the current input information can be obtained.

[0114] Optionally, the preset intelligent agent can generate feedback information based on the current input information and the remembered prompt information, using a preset large language model. It is understandable that combining historical information extracted from past interactions and related to the current input information as memory prompt information can improve the memory capacity of the preset large language model, facilitate the generation of corresponding feedback information based on historical information, improve the accuracy of feedback information, and enhance the user experience.

[0115] The embodiments of this application do not limit the subsequent steps of generating feedback information. Optionally, the feedback information can be displayed to the target user or sent to the target user for their convenience.

[0116] For ease of understanding, this application also provides an application embodiment.

[0117] In this embodiment, an intelligent agent can refer to an artificial intelligence system capable of perceiving the environment, making autonomous decisions, and executing actions to achieve specific goals, possessing autonomy, adaptability, and interactivity. Knowledge retrieval can be the process of organizing and storing information or knowledge in a certain way, and finding relevant information and knowledge according to user needs. A knowledge graph can refer to a technology that presents knowledge in a graphical structure.

[0118] This embodiment proposes a large model memory layer service method and system based on time-aware graphs. The aim is to dynamically fuse unstructured dialogues and structured data through a knowledge graph architecture, preserving their historical relationships and temporal validity. This embodiment addresses the increasing prevalence of intelligent agent applications within banks, particularly the widespread use of knowledge retrieval functions in intelligent agent systems. It can significantly improve the knowledge memory performance of large language models in intelligent agent applications, especially in long dialogues and temporal reasoning tasks.

[0119] This embodiment proposes a method and system for providing a large model memory layer service based on time-aware graphs. The method mainly revolves around the construction of knowledge graphs and the reordering of the retrieval process, obtaining structured results based on agent memory retrieval through data input.

[0120] 1. The overall process may include the following steps: (1) Data ingestion. Receive data input from the dialogue stream and database, and dynamically fuse them. (2) Knowledge graph construction. Construct various knowledge graphs from the input data. (3) Hybrid retrieval and reordering. Use multiple algorithms in parallel to retrieve and reorder data. (4) Context compression and structured construction. Based on the above steps, perform structured processing and compression optimization on the unstructured data of the original dialogue and documents.

[0121] 2. Data Ingestion. Specifically, this may include the following steps: (1) Receiving real-time dialogue streams input from the system, i.e., chat logs between the user and the agent. (2) Data input from the business database, specifically the user's business data. (3) Dynamically fusing the input data and adding two timestamps to each data item, namely the recording timestamp and the actual occurrence timestamp.

[0122] 3. Knowledge Graph Construction. The core of the knowledge graph constructed in this embodiment is to gradually abstract structured knowledge from the original data content through the generation of three-layer subgraphs. (1) Context Subgraph Generation: Used to store the original dialogue messages and retain the complete context, serving as the underlying data source. Nodes are entities or original data content, and edges are context nodes associated with semantic entity nodes. (2) Semantic Entity Subgraph Generation: Used to extract structured knowledge from the context subgraph, that is, to extract the semantic entities in the original data content of the context subgraph nodes and their relationships as relation edges, construct the fact network and merge duplicate semantic entity items. (3) Community Subgraph Generation: Used to cluster the strongly associated entities in the semantic entity subgraph, that is, to aggregate the semantic entity subgraph as the top-level abstraction in the knowledge graph. A community node represents an entity cluster, and the node content includes the community name and summary, key terms, and its embedding vector representation. Using a dynamic label propagation algorithm, new entities are assigned to communities in real time according to relation weights, and communities are reconstructed periodically. The nodes are community nodes and semantic entity nodes, and the edges are the member entity relationships within the community, that is, the affiliation relationships of the semantic entity nodes under the community nodes.

[0123] 4. Semantic Entity Subgraph Generation. In the semantic entity subgraph, nodes are semantic entity nodes, and edges are semantic relationship edges of entities. Specifically, it may include the following steps: (1) Identify entities and relationships in the message through a large language model, and extract relationships and associate entities from the original data content of the context subgraph; (2) Use a large language model to critically generate results and perform conflict detection, with time awareness as the core. If a contradiction is detected, the old fact is marked; (3) Attach an effective time range to each entity relationship.

[0124] 5. Hybrid retrieval and reordering, and contextual structured construction. This step can use a hybrid strategy to filter potential relevant nodes in the constructed hierarchical knowledge graph, query in the order of community subgraph to semantic entity subgraph to context subgraph, and reorder the search results to remove redundancy and sort by importance, resulting in a structured and concise information output. Specifically, it can include the following steps: (1) performing parallel three-way search; (2) taking the union of the three-way results; (3) reordering the results after hybrid search, which can support multiple strategies; (4) converting the filtered graph data into minimal text and formatting the output.

[0125] 6. Parallel three-way search. Specifically, it can include: (1) Semantic search: calculating the cosine similarity between the query and the entity description; (2) Keyword search: matching keywords in the community summary; (3) Graph traversal search: breadth-first search starting from the query entity.

[0126] 7. Reordering. This step mainly prioritizes key information and removes redundancy. Multiple strategies can be used individually or in combination to output a simplified mixed retrieval result for structured information output. Specifically, it can include the following steps: (1) Graph structure reordering is mainly based on the frequency of occurrence in the context dialogue and the distance between nodes: prioritize nodes that appear frequently; take the query-related nodes as the center and prioritize nodes that are close to them; (2) Cross-encoder method, use a large model to directly evaluate the relevance between the query and the result, obtain a relevance score, and sort according to the score.

[0127] 8. Intelligent Agent Flow in Actual Use. The method provided in this embodiment can be embedded into the intelligent agent flow as a large model memory framework. The user inputs content in the intelligent agent flow, which is then processed by the time-aware large model memory layer for information retrieval and context construction, and input into the large language model node for processing, generating output results and returning them to the user. Specifically, it may include the following steps: (1) The user inputs content at the beginning of the intelligent agent flow; (2) The user input is processed through this embodiment, and the context combining the large model memory data and the user input is output; (3) The large language model processes the input; (4) The output results are presented to the user in the intelligent agent flow.

[0128] This embodiment proposes a large model memory layer service method based on time-aware graphs. This method is a knowledge graph engine with time-aware capabilities, which can effectively solve the bottleneck of dynamic data fusion and performs well in long dialogue and temporal reasoning tasks, providing high-throughput and low-latency memory services for large language model agents. This method can also effectively reduce entity extraction illusions, and the time validity processing in fact extraction can automatically invalidate contradictory facts.

[0129] Corresponding to the above method embodiments, this application also provides a user interaction device based on an intelligent agent. The following will be combined with... Figure 3 The device is described in detail.

[0130] Figure 3 A schematic block diagram of a user interaction device based on an agent according to an embodiment of this application is shown.

[0131] like Figure 3 As shown, the user interaction device 300 based on intelligent agents provided in this embodiment includes: a context subgraph module 310, a semantic entity module 320, a community subgraph module 330, and a processing module 340.

[0132] The scenario subgraph module 310 is used to update the scenario subgraph in the current preset knowledge graph based on the current input information of the target user to the preset intelligent agent; the scenario subgraph is used to store the historical interaction information between the target user and the preset intelligent agent. In one embodiment, the scenario subgraph module 310 can be used to perform the operation S210 and related operations described above, which will not be repeated here.

[0133] The semantic entity module 320 is used to update the semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities. Any node in the semantic entity subgraph represents a semantic entity; any edge in the semantic entity subgraph represents the relationships between the semantic entities represented by the different connected nodes. In one embodiment, the semantic entity module 320 can be used to perform the operation S220 described above and related operations, which will not be repeated here.

[0134] The community subgraph module 330 is used to perform semantic entity clustering on the updated semantic entity subgraph and update the community subgraph in the current preset knowledge graph based on the semantic entity clustering results. Any node in the community subgraph represents a semantic entity cluster, and any edge in the community subgraph represents the association between semantic entity clusters represented by different connected nodes. In one embodiment, the community subgraph module 330 can be used to perform the operation S230 described above and related operations, which will not be repeated here.

[0135] The processing module 340 is used to retrieve target nodes related to the current input information in the current preset knowledge graph; determine memory prompt information based on the retrieved target nodes; and generate feedback information for the current input information based on a preset intelligent agent, taking into account the current input information and the memory prompt information. In one embodiment, the processing module 340 can be used to perform the operation S240 described above and related operations, which will not be repeated here.

[0136] Optionally, the scenario subgraph is also used to store interaction context information of different historical interaction information; the scenario subgraph module 310 is used to update the scenario subgraph in the current preset knowledge graph according to the target user's current input information to the preset agent and the interaction context information of the current input information.

[0137] Optionally, the semantic entity module 320 is further configured to: identify semantic entities and the relationships between different semantic entities for the updated context subgraph; perform contradiction detection on the identified relationships; and determine the chronological order of the contradictory relationships when it is determined that contradictory relationships are identified for any two semantic entities. The semantic entity module 320 is configured to: update the semantic entity subgraph in the current preset knowledge graph according to the semantic entities in the updated context subgraph, the relationships between different semantic entities, and the chronological order of the determined relationships.

[0138] Optionally, the processing module 340 is used to: retrieve candidate nodes related to the current input information in the current preset knowledge graph through different retrieval methods to obtain different candidate node sets; and determine the target node related to the current input information based on the union of the different candidate node sets obtained.

[0139] Optionally, the processing module 340 is configured to perform at least one of the following: in the current preset knowledge graph, retrieve target nodes related to the current input information based on the similarity between the current input information and the semantic entity information; in the current preset knowledge graph, retrieve target nodes related to the current input information based on the similarity between the current input information and the semantic entity cluster summary information; the semantic entity cluster summary information is determined based on the information of the semantic entities in the corresponding semantic entity cluster; in the current preset knowledge graph, retrieve target nodes related to the current input information based on the nodes used to represent any semantic entity in the current input information.

[0140] Optionally, the processing module 340 is used to: determine memory prompt information from the retrieved target nodes based on the target nodes whose correlation with the current input information is greater than a preset correlation threshold.

[0141] Optionally, the method for determining the degree of association between the target node and the current input information includes at least one of the following: determining the degree of association between the target node and the current input information based on the frequency of occurrence of the target node in different search results; determining the target node based on different search results obtained from different search methods; determining the degree of association between the target node and the current input information based on the node distance between the target node and the node used to represent the current input information; or determining the degree of association between the target node and the current input information based on a preset large language model. The processing module 340 can be used to determine the degree of association between the target node and the current input information.

[0142] According to embodiments of this application, any plurality of modules among the scenario subgraph module 310, semantic entity module 320, community subgraph module 330, and processing module 340 may be merged into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the scenario subgraph module 310, semantic entity module 320, community subgraph module 330, and processing module 340 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any one of the three implementation methods, or in a suitable combination of any of them. Alternatively, at least one of the scenario subgraph module 310, semantic entity module 320, community subgraph module 330, and processing module 340 may be implemented at least partially as a computer program module that can perform corresponding functions when the computer program module is run.

[0143] For an explanation of this device embodiment, please refer to other embodiments. Each embodiment in the above method embodiment can be executed by the corresponding module in this device embodiment.

[0144] Figure 4 A block diagram schematically illustrates an electronic device suitable for implementing an agent-based user interaction method according to an embodiment of this application.

[0145] like Figure 4 As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0146] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0147] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0148] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0149] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0150] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement any of the method embodiments provided in the embodiments of this application.

[0151] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0152] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0153] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0154] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0155] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0156] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A user interaction method based on intelligent agents, characterized in that, The method includes: Based on the target user's current input information to the preset intelligent agent, update the scenario subgraph in the current preset knowledge graph; the scenario subgraph is used to store the historical interaction information between the target user and the preset intelligent agent. Based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities, the semantic entity subgraph in the current preset knowledge graph is updated; any node in the semantic entity subgraph is used to represent a semantic entity; any side in the semantic entity subgraph is used to represent the relationships between the semantic entities represented by the different nodes connected to it. Semantic entity clustering is performed on the updated semantic entity subgraph, and the community subgraph in the current preset knowledge graph is updated according to the semantic entity clustering results; any node in the community subgraph is used to represent a semantic entity cluster, and any side in the community subgraph is used to represent the association relationship between the semantic entity clusters represented by different connected nodes; In the current preset knowledge graph, target nodes related to the current input information are retrieved; memory prompt information is determined based on the retrieved target nodes; and feedback information is generated for the current input information based on the preset intelligent agent, in response to the current input information and the memory prompt information.

2. The method according to claim 1, characterized in that, The scenario subgraph is also used to store interaction context information for different historical interaction information; The step of updating the scenario subgraph in the current preset knowledge graph based on the target user's current input information to the preset agent includes: Based on the target user's current input information to the preset intelligent agent, and the interaction context information of the current input information, update the scenario subgraph in the current preset knowledge graph.

3. The method according to claim 1, characterized in that, The method further includes: For the updated context subgraph, identify semantic entities and the relationships between different semantic entities; Conflict detection is performed on the identified relationships; if contradictory relationships are identified for any two semantic entities, the chronological order of the contradictory relationships is determined separately. The step of updating the semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph and the relationships between different semantic entities includes: Based on the semantic entities in the updated scenario subgraph, the relationships between different semantic entities, and the temporal order of the determined different relationships, update the semantic entity subgraph in the current preset knowledge graph.

4. The method according to claim 1, characterized in that, The step of retrieving target nodes related to the current input information in the current preset knowledge graph includes: In the current preset knowledge graph, candidate nodes related to the current input information are retrieved using different retrieval methods to obtain different sets of candidate nodes; Based on the union of the different candidate node sets obtained, the target node related to the current input information is determined.

5. The method according to claim 1, characterized in that, The step of retrieving target nodes related to the current input information in the current preset knowledge graph includes at least one of the following: In the current preset knowledge graph, target nodes related to the current input information are retrieved based on the similarity between the current input information and semantic entity information. In the current preset knowledge graph, target nodes related to the current input information are retrieved based on the similarity between the current input information and the semantic entity cluster summary information. The semantic entity cluster summary information is determined based on the information of the semantic entities in the corresponding semantic entity cluster; In the current preset knowledge graph, based on the node used to represent any semantic entity in the current input information, the target node related to the current input information is retrieved.

6. The method according to claim 1, characterized in that, The step of determining memory prompt information based on the retrieved target node includes: Among the retrieved target nodes, memory prompt information is determined based on the target nodes whose correlation with the current input information is greater than a preset correlation threshold.

7. The method according to claim 6, characterized in that, The method for determining the degree of association with the current input information includes at least one of the following: The degree of association between the target node and the current input information is determined based on the frequency of the target node's appearance in different search results; the target node is determined based on different search results obtained using different search methods. The degree of association between the target node and the current input information is determined based on the node distance between the target node and the node used to represent the current input information. Based on a pre-defined large language model, the degree of association between the target node and the current input information is determined.

8. A user interaction device based on an intelligent agent, characterized in that, The device includes: The scenario subgraph module is used to update the scenario subgraph in the current preset knowledge graph based on the current input information of the target user to the preset intelligent agent; the scenario subgraph is used to store the historical interaction information between the target user and the preset intelligent agent; The semantic entity module is used to update the semantic entity subgraph in the current preset knowledge graph based on the semantic entities in the updated scenario subgraph and the association relationships between different semantic entities; any node in the semantic entity subgraph is used to represent a semantic entity; any side in the semantic entity subgraph is used to represent the association relationship between the semantic entities represented by the different nodes connected to it; The community subgraph module is used to perform semantic entity clustering on the updated semantic entity subgraph and update the community subgraph in the current preset knowledge graph based on the semantic entity clustering results; any node in the community subgraph is used to represent a semantic entity cluster, and any side in the community subgraph is used to represent the association relationship between the semantic entity clusters represented by the different connected nodes. The processing module is used to retrieve target nodes related to the current input information in the current preset knowledge graph; determine memory prompt information based on the retrieved target nodes; and generate feedback information for the current input information based on the preset intelligent agent in response to the current input information and the memory prompt information.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images