Information updating method and device based on large model, agent and equipment

Abstract

The disclosure provides a large model-based information updating method and device, an agent and equipment, relates to the technical field of artificial intelligence, and in particular to the fields of intelligent interaction, information technology and the like. The specific implementation scheme of the large model-based information updating method is as follows: a large model is used to analyze historical information associated with a target task to obtain to-be-operated information; the historical information is obtained based on the task requirements of the target task during the execution of the target task by the agent; the to-be-operated information indicates an operation type and target information of an update operation performed on a to-be-updated object in the historical information; there is a difference between the historical content information of the to-be-updated object and input information used to execute the target task; and the historical content information of the to-be-updated object is updated to the target information by performing the update operation on the to-be-updated object according to the operation type.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to fields such as intelligent interaction and information technology, and specifically to information updating methods, devices, intelligent agents, and equipment based on large models. Background Technology

[0002] In intelligent interaction scenarios, intelligent agents can build long-term memory information, such as user preferences and reusable experiences, by performing multiple tasks, so that they can be directly recalled in subsequent tasks.

[0003] However, due to the limited context window of the agent, as the complexity of intelligent interaction scenarios increases, various outdated or erroneous memory information may continue to affect the agent's output results, thus impacting the accuracy of the agent's output results. Summary of the Invention

[0004] This disclosure provides a method, apparatus, intelligent agent, and device for information updating based on a large model.

[0005] According to one aspect of this disclosure, a method for updating information based on a large model is provided, comprising: analyzing historical information associated with a target task using a large model to obtain information to be operated on; wherein the historical information is obtained by the agent based on the task requirements of the target task during the execution of the target task; the information to be operated on indicates the operation type and target information for performing an update operation on the object to be updated in the historical information; there is a difference between the historical content information of the object to be updated and the input information used to perform the target task; and updating the historical content information of the object to be updated to the target information by performing an update operation on the object to be updated according to the operation type.

[0006] According to another aspect of this disclosure, an information updating device based on a large model is provided, comprising: an analysis module and an execution module.

[0007] The analysis module is used to analyze historical information associated with the target task using a large model to obtain information to be operated on. The historical information is obtained by the agent based on the task requirements of the target task during the execution of the target task. The information to be operated on indicates the operation type and target information for performing update operations on the object to be updated in the historical information. There is a difference between the historical content information of the object to be updated and the input information used to execute the target task.

[0008] The execution module is used to update the historical content information of the object to be updated to the target information by performing update operations on the object to be updated according to the operation type.

[0009] According to embodiments of this disclosure, an intelligent agent is provided, including: an input module, a processing module, and an output module.

[0010] The input module is used to receive historical information associated with the target task; wherein, the historical information is obtained by another intelligent agent during the execution of the target task based on the task requirements of the target task.

[0011] The processing module is used to determine the target task based on the historical information received by the input module, determine the target large model based on the target task, and execute the information update method based on the large model described above by calling the target large model to obtain the operation information to be operated; the operation information to be operated indicates the operation type and target information for performing the update operation on the object to be updated in the historical information; there is a difference between the historical content information of the object to be updated and the input information used to perform the target task.

[0012] The output module is used to output the information to be operated obtained by the processing module, so that the historical content information of the object to be updated can be updated to the target information by performing an update operation on the object to be updated according to the operation type.

[0013] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0014] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the methods described above.

[0015] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described above.

[0016] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0018] Figure 1 This illustration schematically shows an exemplary system architecture for applying a large-model-based information update method and apparatus according to embodiments of the present disclosure;

[0019] Figure 2 A flowchart illustrating an information update method based on a large model according to an embodiment of the present disclosure is shown schematically.

[0020] Figure 3A This diagram illustrates an embodiment of the present disclosure of obtaining information to be operated.

[0021] Figure 3B A schematic diagram illustrating the acquisition of information to be operated according to another embodiment of the present disclosure is shown.

[0022] Figure 4 This illustration schematically shows an update diagram for similar semantic information according to an embodiment of the present disclosure;

[0023] Figure 5 This illustration schematically shows an update diagram for contradictory semantic information according to an embodiment of the present disclosure;

[0024] Figure 6 This illustration schematically depicts a phased recall of information to execute a task according to an embodiment of the present disclosure;

[0025] Figure 7 A block diagram of a large-model-based information updating apparatus according to an embodiment of the present disclosure is shown schematically.

[0026] Figure 8 A block diagram of an intelligent agent according to an embodiment of the present disclosure is schematically shown; and

[0027] Figure 9 A block diagram of an electronic device suitable for implementing a large-model-based information update method according to an embodiment of the present disclosure is illustrated. Detailed Implementation

[0028] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0029] The agent needs to remember long-term user preferences, identity profiles, business facts, operational constraints, reusable experiences, and important conversation summaries between multiple tasks, and use them as needed in subsequent tasks.

[0030] Examples typically involve directly appending historical dialogues or summaries to the next round of prompts; preference files, rule files, or memo files with fixed formats are maintained by the agent or user. Historical fragments are written to a full-text index or vector library and recalled based on similarity at the start of a new task.

[0031] However, the context window of an intelligent agent is limited, and historical information includes temporary task states, expired facts, sensitive information, duplicate content, and information from different scopes. Simply saving or recalling information can easily lead to problems such as memory pollution, unauthorized use, accidental writing, expired information continuing to affect decisions, and increased system latency.

[0032] In view of this, the embodiments of this disclosure utilize the natural language understanding and logical reasoning capabilities of a large model to analyze historical information associated with the target task. From this historical information, objects that differ from the input information used to perform the target task are selected and need to be updated, generating information to be operated on. Compared to the method in related examples of directly concatenating historical dialogues or summaries into prompts, utilizing the natural language understanding capabilities of a large model more accurately identifies and filters historical information that needs updating. This reduces the long-term injection of irrelevant, expired, or undesirable historical information into the context of subsequent interactions. It enables the agent to stably inherit relatively accurate and reusable historical information in the interaction scenario, reduces the interference of redundant or erroneous historical information on the agent's execution decisions, and further improves the accuracy and efficiency of the agent's execution results.

[0033] Figure 1 The illustration schematically shows an exemplary system architecture for applying large-model-based information update methods and apparatus according to embodiments of the present disclosure.

[0034] It is important to note that Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments, or scenarios. For example, in another embodiment, an exemplary system architecture for applying the information update method and apparatus based on a large model may include a terminal device, but the terminal device can implement the information update method and apparatus based on a large model provided by the embodiments of this disclosure without interacting with the server.

[0035] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a database 101, an agent 102, and a large model 103.

[0036] Database 101 can be used to store information generated by the intelligent agent 102 during multiple interactions with the user in historical phases, which can be reused in subsequent interactions. This information can be called memory information. Examples include: long-term user preferences, identity profiles, business facts, operational constraints, reusable experiences, and important session summaries.

[0037] The intelligent agent 102 can retrieve historical information related to the task requirements of the task to be executed from the database 101 based on the task to be executed. This historical information can record reusable experience information related to the task to be executed, such as user profiles and operational constraints.

[0038] The intelligent agent 102 can execute the task based on the acquired historical information and the task requirements of the task to be executed, and output the result information.

[0039] The large model 103 can acquire historical information and task information of tasks to be executed by the agent 102 during task execution, or after the agent 102 has completed the task. Based on the historical information and task information, it generates operation information to be performed. Based on the operation information, it performs update operations on the objects to be updated in the historical information in the database 101. This enables the dynamic updating of the memory information stored in the database 101 according to the various tasks to be performed during the continuous interaction between the agent 102 and the user.

[0040] It should be understood that Figure 1 The number of large models, agents, and databases shown is merely illustrative. Depending on implementation needs, there can be any number of large models, agents, and databases.

[0041] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of any type of information, such as user personal information, comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0042] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0043] Figure 2 A flowchart illustrating an information update method based on a large model according to an embodiment of the present disclosure is shown schematically.

[0044] like Figure 2 As shown, the information update method 200 based on a large model may include operations S210 to S220.

[0045] In operation S210, the historical information associated with the target task is analyzed using a large model to obtain the information to be operated.

[0046] In operation S220, according to the operation type, the historical content information of the object to be updated is updated to the target information by performing an update operation on the object to be updated.

[0047] Historical information is acquired by the intelligent agent during the execution of the target task, based on the task requirements of the target task. Intelligent agents can include, but are not limited to, conversational intelligent agents, task-oriented intelligent agents, office intelligent agents, research and development intelligent agents, and other intelligent agents with multi-turn interaction, tool invocation, and cross-task execution capabilities.

[0048] Historical information can include reusable information recorded by an agent during multiple executions of historical tasks over a historical period; it can also be called memory information. This reusable information includes, but is not limited to, long-term user preferences, identity profiles, business facts, operational constraints, reusable experiences, and summaries of important sessions.

[0049] In some embodiments, historical information and prompts can be input into the large model. The prompts may include contextual information about the target task, such as the source of the target task, its input information, and its processing results. The prompts may also include candidate object information, which can refer to objects in the historical information that can be updated, such as user preferences. By leveraging the large model's natural language understanding and logical reasoning capabilities, combined with the context of the target task, objects to be updated that differ from the input information used to perform the target task can be filtered from the historical information.

[0050] The object to be updated can represent structured data that records various types of reusable information. The historical content information of this structured data can include at least the memory content, memory type, scope, source, confidence level, salience, vitality value, lifecycle status, version number, update time, and audit identifier.

[0051] Scope limits the scope of reuse scenarios for reusable information. For example, memory data from project group A cannot be used in pending tasks triggered by individual users, thereby reducing the probability of unauthorized use and improving data security. Memory types can include user profiles, long-term preferences, behavioral constraints, business facts, operational experience, and session summaries.

[0052] Vitality or validity status can be adjusted based on access frequency, last used time, explicit user confirmation, fact updates, negative feedback, and type weights to achieve lifecycle management of memory information. For example, high-value memories can be fixed or prioritized; memories that have not been used for a long time, have low confidence, or have been replaced by new facts can be decayed, archived, or deleted; sensitive or unauthorized memories can be immediately isolated. By adjusting vitality or validity status, the memory information stored in the database can be continuously updated with usage, rather than just being appended without being cleaned up.

[0053] By transforming reusable information into structured data, memory is no longer uncontrollable free text, but rather a filterable, sortable, updatable, and interpretable management unit.

[0054] The pending operation information indicates the type of operation and target information for performing an update operation on the object to be updated in the historical information. The operation type can include, but is not limited to, operations such as adding, updating, deleting, replacing, and merging. The target information can represent the updated content obtained after performing the update operation on the object to be updated.

[0055] The target information can be extracted by the large model from the input information used to perform the target task, or it can be extracted from the execution result information of the target task, or it can be generated by combining at least two of the input information, historical information and execution result information through comprehensive analysis.

[0056] In some embodiments, the information to be operated on may further include the reason for performing the update operation on the object to be operated on, and supporting evidence information. This reason for update and supporting evidence information can serve as criteria for the user to determine whether to trigger the update operation on the object to be updated. For example, the user's judgment can be used to determine whether to trigger the update operation on the object to be updated.

[0057] In some embodiments, the historical content information of the object to be updated can be updated to the target information by performing an update operation on the object to be updated, according to the operation type. For example, if the historical content information of the object to be updated records "A likes sports", the operation type indicated by the operation information is "add operation", and the target information can be "A likes outdoor sports", then by performing an update operation, "A likes sports" can be updated to "A likes outdoor sports". Through memory update, the intelligent agent can more accurately perform subsequent interactive tasks.

[0058] This embodiment utilizes the natural language understanding and logical reasoning capabilities of a large-scale model to analyze historical information associated with the target task. It then filters out objects from this historical information that differ from the input information used to perform the target task, generating information to be updated. Compared to related examples that directly append historical dialogues or summaries to prompts, this method, leveraging the natural language understanding capabilities of a large-scale model, more accurately identifies and filters out historical information that needs updating. This reduces the long-term injection of irrelevant, outdated, or undesirable historical information into the context of subsequent interactions. It enables the agent to stably inherit relatively accurate and reusable historical information in interactive scenarios, reducing the interference of redundant or erroneous historical information on the agent's execution decisions, and further improving the accuracy and efficiency of the agent's execution results.

[0059] The following is combined Figure 3A and Figure 3B The specific embodiments for obtaining the information to be operated according to the present disclosure will be described in detail.

[0060] Figure 3AA schematic diagram illustrating the acquisition of information to be operated according to an embodiment of the present disclosure is shown.

[0061] like Figure 3A As shown, the large model is used to analyze the historical information associated with the target task to obtain the information to be operated, which may include the following operations: using the large model 302 to analyze the historical information 301 associated with the target task to obtain candidate operation information; and determining the information to be operated from the candidate operation information.

[0062] Candidate operation information indicates the type of candidate operation to be performed on each of the multiple candidate objects in the historical information. Since the candidate operation information is generated by the large model based on logical reasoning capabilities, its content has a certain degree of uncontrollable probability. To reduce erroneous or invalid updates to historical information, a verification mechanism for the candidate operation information can be introduced, allowing update operations to be performed on candidate objects that pass the verification. Then, the candidate operation information that passes the verification is determined as the information to be operated on.

[0063] In some embodiments, candidate operation information can be validated by configuring preset rules. Validation content may include, but is not limited to, permission checks, duplicate checks, and whether the updated content is "empty." Permission checks can verify whether the initiating object of the current target task has permission to modify the historical content information of the object to be updated, thereby reducing the probability of unauthorized updates to historical information and further improving data security. Duplicate checks can verify whether the target information and historical content information are different in description but have the same actual semantics.

[0064] For example: Figure 3A Although the candidate operation information C2 indicates the operation type as "add", the content to be added is "empty". Therefore, it can be determined that the operation information to be performed is the candidate operation information C1 "delete memory ID1".

[0065] Figure 3A The Prompt shown can represent a prompt message, which may include context information about the target task. This context information may record the identifier of the candidate operation object, such as memory ID1.

[0066] In some embodiments, candidate objects existing in the context information of the target task can also be determined from the candidate operation information as objects to be updated based on the context information of the target task.

[0067] The candidate objects indicated by the candidate operation information can be the same as or different from the candidate operation objects indicated in the prompt information. This is because during the logical reasoning process, the large model may generate objects that were not originally present in the prompt information; these are invalid objects to be updated. Therefore, the operation information can be determined from the candidate operation information by matching the identifiers of the candidate operation objects.

[0068] For candidate operation information generated by large models, the information to be operated is determined from the candidate operation information by means of preset rules or identifier matching, which reduces the probability of performing erroneous or invalid update operations on the object to be updated and further improves the information update efficiency.

[0069] Figure 3B A schematic diagram illustrating the acquisition of information to be operated according to another embodiment of the present disclosure is shown.

[0070] In addition to candidate object identifiers and candidate operation types, candidate operation information can also include interaction scenario information applicable to each of the multiple candidate objects. For example... Figure 3B As shown, candidate operation information C1 may include "delete memory ID1 in the discussion group of project A", and candidate operation information C2 may include "merge memory ID1 and memory ID6 in the discussion group of project A".

[0071] exist Figure 3B In the illustrated embodiment, the specific method for generating candidate operation information using a large model is the same as described above. Figure 3A The same applies to the embodiments shown, and will not be repeated here.

[0072] According to embodiments of this disclosure, determining the information to be operated from candidate operation information may include the following operations: determining the object to be updated that matches historical information from a plurality of candidate objects; and determining the operation type associated with updating the object to be updated as the information to be operated in response to determining that the interaction scenario information applicable to the object to be updated matches the execution environment information of the target task.

[0073] like Figure 3B As shown, we can first filter based on the historical information "Memory ID2 and ID1 in Project A discussion group". Since the historical objects involved in the historical information only include Memory ID2 and ID1, it can be understood that Memory ID2 and ID1 are real stored structured data objects. However, Memory ID6 mentioned in the candidate operation information C2 is a fabricated data object created during the generation of the large model. Therefore, the object to be updated that matches the historical information is "Memory ID1".

[0074] In addition to filtering invalid data objects, to improve data security, this disclosure also introduces interaction scenario information applicable to structured data. This interaction scenario information can be used to distinguish the permission scope between different types of structured data. For example, interaction scenario information may include personal interaction scenarios, group interaction scenarios, project interaction scenarios, etc. In a group interaction scenario, the update permission for structured data can be limited to some members within the group, while in a project interaction scenario, the update permission for structured data can be limited to all or some members responsible for the project.

[0075] like Figure 3B As shown, the memory ID6 mentioned in candidate operation information C2 is an invalid data object; therefore, there is no need to verify candidate operation information C2 again. It is only necessary to verify whether the interaction scenario information "Project A discussion group" applicable to the object to be updated "Memory ID1" mentioned in candidate operation information C1 matches the target task execution environment information. The target task execution environment information can refer to the interaction scenario information used to trigger the target task, such as "Project A discussion group". Through matching verification, it can be determined that the operation information to be performed is "Delete Project A discussion group memory ID1".

[0076] In some embodiments, when the interaction scenario information applicable to the object to be updated in the candidate operation information does not match the target task execution environment information, the candidate operation information can also be sent to the user to determine whether to perform an update operation on the object to be updated, or to cover or narrow down the interaction scenario information applicable to the updated object, based on user feedback. For example, if the interaction scenario applicable to the object to be updated indicated in the candidate operation information is project group A, the interaction scenario applicable to the object to be updated can be narrowed down to user B based on user feedback.

[0077] By verifying whether the interaction scenario information applicable to the object to be updated in the candidate operation information matches the execution environment information of the target task, mutual isolation between memory information generated under different interaction scenarios is achieved, which further improves the data security of memory information generated under different interaction scenarios and reduces the risk of unauthorized use or unauthorized updates.

[0078] In real-world scenarios, the operational information generated by the large model can also include expected updated information for the object to be updated. This expected updated information can be extracted by the large model from the input information used to perform the target task, or from the execution result information of the target task, or it can be generated by comprehensively analyzing at least two of the input information, historical information, and execution result information. This expected updated information may conflict with or overlap with the historical content information of the object to be updated, affecting the efficiency of information updating.

[0079] In view of this, the method provided in the embodiments of this disclosure may further include the following operations: in response to determining that the historical content information of the object to be updated and the expected updated information are each applicable to the same interactive scenario information, performing semantic analysis on the historical content information and the expected updated information of the object to be updated to obtain semantic similarity; and generating target information based on the semantic similarity using a large model based on the historical content information and the expected updated information.

[0080] For example, the interaction scenario information applicable to the historical content information of the object to be updated can be "Discussion Group A of Project Group", and the interaction scenario information applicable to the expected updated information can be "Discussion Group A of Project Group". It can be determined that the interaction scenario information applicable to the historical content information of the object to be updated and the expected updated information are the same.

[0081] At this point, any semantic similarity algorithm can be used to calculate the semantic similarity between the historical content information and the expected updated information of the object to be updated, or a large model can be called to perform semantic analysis on the historical content information and the expected updated information of the object to be updated to obtain the semantic similarity.

[0082] For example, historical content information could be "I am a surfing enthusiast," while the expected updated information could be "I like surfing." Although the textual descriptions of the historical content information and the expected updated information are different, the semantics they express are the same. Therefore, a large model can be used to generate target information based on semantic similarity. For example, the target information could be "I enjoy surfing."

[0083] By combining the historical content information of the object to be updated with the expected updated information generated by the large model, semantic similarity is judged, which reduces redundant update operations performed on information that is semantically similar but has different content, and further improves the efficiency of information update.

[0084] In real-world interaction scenarios, especially for accumulated user preference information, users have a high degree of freedom in expressing themselves. There may be an inclusion relationship between the semantics expressed by historical content information and the expected updated information.

[0085] According to embodiments of this disclosure, generating target information based on semantic similarity using a large model based on historical content information and expected updated information may include the following operations: in response to determining that the semantic similarity is less than a predetermined threshold, semantically merging the historical content information and expected updated information using a large model to generate target information.

[0086] Figure 4 The illustration shows a schematic diagram of updating similar semantic information according to an embodiment of the present disclosure.

[0087] like Figure 4As shown, the target task can be triggered during the interaction between user A and the intelligent agent, and the object to be updated can be the memory ID2 associated with user A in the database. The historical content information of memory ID2 can be "XX likes fruit", and the expected updated information of memory ID2 can be "XX likes apples".

[0088] By calculating semantic similarity or semantic analysis, it can be determined that the semantic similarity between the historical content information of memory ID2 and the expected updated information of memory ID2 is less than a predetermined threshold. However, since apples are fruits, it can be determined that the expected updated information describes user preferences more accurately. Therefore, the historical content information and the expected updated information can be semantically merged to obtain the target information as "XX likes fruits, especially apples".

[0089] Then, an information update operation can be performed on memory ID2 associated with user A in the database, changing the content of memory ID2 from "XX likes fruit" to "XX likes fruit, especially apples". This preserves the original historical memory and further improves the accuracy of the memory information without narrowing the query scope for subsequent tasks to obtain the required information from the historical information.

[0090] According to embodiments of this disclosure, generating target information based on historical content information and expected updated information using a large model based on semantic similarity may include the following operations: in response to determining that the semantic similarity is greater than or equal to a predetermined threshold, performing semantic analysis on the historical content information and expected updated information using a large model to obtain analysis results; and in response to determining that the analysis results indicate that there is a semantic contradiction between the historical content information and the expected updated information, determining the expected updated information as the target information.

[0091] Figure 5 The illustration shows a schematic diagram of updating contradictory semantic information according to an embodiment of the present disclosure.

[0092] like Figure 5 As shown, the target task can be triggered during the interaction between users and agents in the discussion group of Project A, and the object to be updated can be the memory ID2 associated with Project A in the database. The historical information of memory ID2 can be "Location A is located in Zone B", and the expected updated information of memory ID2 can be "Location A is located in Zone C".

[0093] By calculating semantic similarity or performing semantic analysis, it can be determined that the semantic similarity between the historical content information of memory ID2 and the expected updated information of memory ID2 is greater than or equal to a predetermined threshold. However, since “area B” and “area A” have completely different semantics, but both represent the location of the same object “location A”, a large model can be called to perform semantic analysis on the historical content information and the expected updated information. For example, it can be analyzed whether “area B” and “area C” are different names for the same location, or whether “location A” includes two different partition locations, etc., to obtain the analysis results.

[0094] The analysis results indicate whether there is a semantic contradiction between historical content information and the expected updated information. Semantic contradictions in this embodiment can characterize factual contradictions. For example, if location A has only one partition location, and "Area B" and "Area C" are two different areas located in different locations, then location A cannot simultaneously be located in both "Area B" and "Area C." This constitutes a factual contradiction.

[0095] In cases where there is a semantic contradiction between the information and historical content, the expected updated information can be identified as the target information.

[0096] In some embodiments, the content of memory ID2 associated with project A in the database can be updated directly based on the target information. Semantic contradictions between historical content information and expected updated information can also be marked for the object to be updated. The update process and reasons for memory ID2 can be recorded with minimal storage cost so as to perform data traceability analysis on the execution process of subsequent tasks that call memory ID2.

[0097] To address semantic discrepancies between historical information and expected updated information, the expected updated information is prioritized to promptly clear erroneous memories from historical information. This enables dynamic updating and maintenance of memory information, reduces the continued impact of erroneous memories on subsequent interactive tasks, and further improves the accuracy of the output results of subsequent interactive tasks.

[0098] To reduce the impact of information update operations on agent interaction, according to the operation type, updating the historical content information of the object to be updated to the target information by performing update operations on the object to be updated can include the following operations: determining the information of the update task based on the target information and the operation type information; and performing update operations on multiple objects to be updated separately by executing update tasks asynchronously with the target task.

[0099] The information for an update task can include the specific operations to be performed on each of the multiple objects to be updated. This information can be obtained after the agent has engaged in multiple interactions. For example, the update task information could include: performing a deletion operation on memory ID1, adding "data D" to memory ID2, etc.

[0100] By executing update tasks asynchronously with the target task, update operations are performed on multiple objects to be updated, so as to update the historical content information of each of the multiple objects to be updated to their respective target information.

[0101] In some embodiments, any of the operations described above may fail, such as: generating information to be operated on by the large model, performing verification operations on candidate operation information, semantic similarity calculation operations, semantic merging operations, etc. When an abnormal operation occurs, an asynchronous supplementary process can be triggered, such as: asynchronously calling the large model to re-execute the generation operation without affecting the normal interaction of the agent, asynchronously calling the verification tool to perform the verification operation, etc. At the same time, when the asynchronous supplementary process fails, the normal interaction of the agent is not blocked. The agent's interaction information during the execution of the asynchronous supplementary process can be recorded so that the information update operation can be re-executed based on the interaction information during that period.

[0102] By executing information update operations asynchronously, the impact of the computing resources consumed by information update operations on the resources required for agent interaction is reduced, ensuring the response efficiency of agent interaction in scenarios with limited computing resources.

[0103] Figure 6 The illustration schematically depicts a phased recall of information to perform a task according to an embodiment of the present disclosure.

[0104] According to embodiments of this disclosure, the method may further include the following operations: obtaining first historical information associated with the task requirements of the task to be executed from candidate information; and using an intelligent agent to execute the task to be executed based on information concatenated from the first historical information and the task requirements information; wherein the candidate information includes at least one target information.

[0105] Candidate information may include at least one target information, and candidate information may also include unupdated historical information, that is, memory information that has not been corrected in the interaction scenario.

[0106] The trigger time of the task to be executed can be later than the trigger time of the target task described above, or later than the execution time of the change task described above. It can be understood that when the task to be executed is triggered, the historical information in the database has already completed the update operation.

[0107] The task requirements for the task to be executed may include, but are not limited to, task intent, session environment, and current interaction scenario information.

[0108] like Figure 6 As shown, the first historical information can be retrieved from the memory units in the database that match the aforementioned interaction scenario based on the task requirements. The first historical information can be memory information that satisfies the task intent and conforms to the current interaction scenario information and session environment.

[0109] Then, the information obtained by combining the first historical information with the information required by the task can be input into the intelligent agent, so that the intelligent agent can combine the historical experience recorded in the first historical information and execute the task to be executed according to the task requirements.

[0110] Since the candidate information has been dynamically updated according to the large model-based information update method described above, the effectiveness of the historical experience recorded by the first historical information is further improved, the interference of erroneous or redundant memory on the agent is reduced, and the task execution efficiency of the agent is further improved.

[0111] Furthermore, the execution of the task to be performed using the information concatenated by the intelligent agent based on the first historical information and the information of the task requirements may include the following operations: generating information to be retrieved using the information concatenated by the intelligent agent based on the first historical information and the information of the task requirements; obtaining second historical information associated with the information to be retrieved from the candidate information; and executing the task to be performed using the intelligent agent based on the first historical information, the second historical information, and the information of the task requirements.

[0112] like Figure 6 As shown, the agent, based on task requirements and historical experience, determines the historical information needed to execute the task during the task execution process, and generates retrieval information to obtain this historical information. The retrieval information may include, but is not limited to, keywords, semantic vectors, rule tags, or a combination of at least two of these.

[0113] Secondary historical information related to the information to be retrieved is obtained from candidate information. This secondary historical information may include, but is not limited to, hard constraints, identity profiles, user preferences, business facts, operational experience, and summary background. It may also be obtained by preprocessing the recalled data according to pre-set type priority, recall data volume limit, deduplication rules, and expiration suppression rules.

[0114] Finally, the agent executes the task based on the first historical information, the second historical information, and the task requirements. This reduces context bloat caused by simple concatenation and allows for priority constraints on agent behavior, further improving the stability of the agent's output.

[0115] In some embodiments, after updating historical information and completing the pending tasks, the memory source, update reason, verification result, version change, recall hit, and injection location of each structured data in the database can be recorded to trace the memory update process. In some embodiments, the screening results of candidate operation information generated by the large model can also be recorded, allowing for timely adjustments to the update direction for memory information with high recall noise. This further improves the efficiency of updating historical information.

[0116] Figure 7 A block diagram of a large-model-based information updating apparatus according to an embodiment of the present disclosure is shown schematically.

[0117] like Figure 7 As shown, the information update device 700 based on a large model may include an analysis module 710 and an execution module 720.

[0118] Analysis module 710 is used to analyze historical information associated with the target task using a large model to obtain information to be operated; wherein, the historical information is obtained by the agent based on the task requirements of the target task during the execution of the target task; the information to be operated indicates the operation type and target information for performing update operations on the object to be updated in the historical information; there is a difference between the historical content information of the object to be updated and the input information used to execute the target task.

[0119] The execution module 720 is used to update the historical content information of the object to be updated to the target information by performing an update operation on the object to be updated according to the operation type.

[0120] According to embodiments of this disclosure, the analysis module 710 may include an analysis submodule and a determination submodule.

[0121] The analysis submodule is used to analyze historical information associated with the target task using a large model to obtain candidate operation information. The candidate operation information indicates the candidate operation type for performing update operations on multiple candidate objects in the historical information.

[0122] The determination submodule is used to determine the operation information from the candidate operation information.

[0123] According to embodiments of this disclosure, the candidate operation information further includes interaction scenario information applicable to each of the multiple candidate objects. The determination submodule may include a matching unit and a determination unit.

[0124] The matching unit is used to identify the object to be updated from multiple candidate objects that matches historical information.

[0125] The determining unit is used to determine, in response to matching the interaction scenario information applicable to the object to be updated with the execution environment information of the target task, to determine that the operation type associated with updating the object to be updated is the operation information to be operated.

[0126] According to embodiments of this disclosure, the information to be operated includes expected updated information for the object to be updated; the aforementioned information update device 700 based on a large model may further include a first semantic analysis module and a generation module.

[0127] The first semantic analysis module is used to perform semantic analysis on the historical content information and the expected updated information of the object to be updated in response to the determination that the historical content information and the expected updated information of the object to be updated are the same in terms of the applicable interaction scenario information, and to obtain the semantic similarity.

[0128] The generation module is used to generate target information based on semantic similarity, utilizing a large model based on historical content information and expected updated information.

[0129] According to embodiments of this disclosure, the generation module may include a first generation submodule, configured to, in response to determining that the semantic similarity is less than a predetermined threshold, semantically merge historical content information and expected updated information using a large model to generate target information.

[0130] According to embodiments of this disclosure, the generation module may include a second semantic analysis submodule and a second generation submodule.

[0131] The second semantic analysis submodule is used to perform semantic analysis on historical content information and expected updated information using a large model in response to determining that the semantic similarity is greater than or equal to a predetermined threshold, and obtain the analysis results; wherein, the analysis results indicate whether there is a semantic contradiction between historical content information and expected updated information.

[0132] The second generation submodule is used to determine the expected updated information as the target information in response to a semantic contradiction between the historical content information indicated by the analysis results and the expected updated information.

[0133] According to embodiments of this disclosure, the execution module includes a task determination submodule and an update submodule.

[0134] The task determination submodule is used to determine the information of the update task based on the target information and the operation type information; the update task information includes the operation information for each of the multiple objects to be updated.

[0135] The update submodule is used to perform update operations on multiple objects to be updated by executing update tasks asynchronously with the target task.

[0136] According to embodiments of this disclosure, the above-described information update device 700 based on a large model may further include an acquisition module and a task execution module.

[0137] The acquisition module is used to retrieve the first historical information associated with the task requirements of the task to be executed from the candidate information.

[0138] The task execution module is used to execute the task to be executed by an intelligent agent based on the information concatenated from the first historical information and the information of the task requirements; wherein, the candidate information includes at least one target information.

[0139] According to embodiments of this disclosure, the task execution module may include a generation submodule, a retrieval submodule, and an execution submodule.

[0140] The generation submodule is used to generate the information to be retrieved by using the information spliced ​​by the agent based on the first historical information and the information required by the task.

[0141] The retrieval submodule is used to retrieve second historical information associated with the information to be retrieved from the candidate information.

[0142] The execution submodule is used to execute the task to be executed by the agent based on the first historical information, the second historical information, and the task requirements.

[0143] According to embodiments of this disclosure, this disclosure also provides an intelligent agent, an electronic device, a readable storage medium, and a computer program product.

[0144] According to embodiments of this disclosure, an intelligent agent includes: an input module, a processing module, and an output module.

[0145] The input module is used to receive historical information associated with the target task; wherein, the historical information is obtained by another intelligent agent during the execution of the target task based on the task requirements of the target task.

[0146] The processing module is used to determine the target task based on the historical information received by the input module, determine the target large model based on the target task, and obtain the information to be operated by calling the target large model to execute the information update method based on the large model described above. The information to be operated indicates the operation type and target information for performing the update operation on the object to be updated in the historical information. There is a difference between the historical content information of the object to be updated and the input information used to perform the target task.

[0147] The output module is used to output the information to be operated obtained by the processing module, so that the historical content information of the object to be updated can be updated to the target information by performing an update operation on the object to be updated according to the operation type.

[0148] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described above.

[0149] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described above.

[0150] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0151] Figure 8 A block diagram of an intelligent agent according to an embodiment of the present disclosure is shown schematically.

[0152] like Figure 8 As shown, in the embodiments of this disclosure, inspired by the von Neumann architecture in modern computer theory, such as... Figure 8 As shown, the AI ​​agent 800 may include three core modules: an input module 810, an output module 820, and a processing module 830. The processing module 830 may include a control unit 831, a storage unit 832, and a computing unit 833.

[0153] The input module 810 is responsible for receiving or sensing information such as queries, requests, instructions, signals, or data from the outside world (e.g., users or the external environment), and converting it into a format that the AI ​​agent 800 can understand and process. The input module 810 is the primary link for the AI ​​agent 800 to interact with the outside world. It enables the AI ​​agent 800 to efficiently and accurately obtain the necessary "sensory" information from the outside world and respond to this information.

[0154] In the example, the input information received by input module 810 can be historical information associated with the target task. This historical information is acquired by another agent during the execution of the target task, based on the task requirements of that task.

[0155] In the example, the processing module 830 is the core support for the AI ​​agent 800's ability to handle complex tasks. The processing module 830 can determine the target task based on the input information received by the input module 810, determine the large model based on the target task, and execute the information update method based on the large model described above by calling the large model, and output the information to be operated.

[0156] In the example, the control unit 831 in the processing module 830 will continuously interact with the storage unit 832, the arithmetic unit 833, and / or the output module 820 during operation. However, it should be noted that in the embodiments of this disclosure, the control unit 831 initiates communication with the storage unit 832, the arithmetic unit 833, and / or the output module 820 as a single initiator, and there is no communication coupling between the storage unit 832, the arithmetic unit 833, and the output module 820.

[0157] In the example, the performance of the control unit 831 is closely related to the large model on which the AI ​​agent 800 is based. To fully leverage the capabilities of the large language model, the internal structure of the control unit 831 can be designed to be highly configurable and scalable to handle various types of tasks and requirements in real-world scenarios.

[0158] Storage unit 832 can be responsible for remembering information such as historical dialogues and event streams. Configuration information, target text, and data resources generated in each round can be included in storage unit 832.

[0159] In the example, after receiving a configuration generation request, the AI ​​agent 800 can determine the configuration intent from the initial text using an intent recognition model. The configuration intent can be stored in storage unit 832. The AI ​​agent 800 can retrieve relevant data resources from storage unit 832 and feed them back to control unit 831. Then, control unit 831 can use the returned data resources to obtain configuration data corresponding to the initial text. It can also retrieve relevant text data from storage unit 832 and feed it back to control unit 831. Then, control unit 831 can use the returned text data to obtain the target text and pass the target text and configuration data to output module 820.

[0160] The arithmetic unit 833 can be viewed as a predefined tool library. Renderers and display controls, as mentioned earlier, can be included in the arithmetic unit 833.

[0161] In the example, when the AI ​​agent 800 needs to render multiple output data, it can call the relevant renderer and display controls from the computing unit 833 and feed them back to the control unit 832. Then, the control unit 832 can use the fed-back renderer and display controls to render the first search result and pass it to the output module 820. It is understandable that although large language models have excellent language understanding and generation capabilities, like humans, the tasks they can solve without any tools are very limited. When the AI ​​agent 800 is given the ability to call tools, it can perform tasks such as using a calculator to complete mathematical calculations, using Python to perform data analysis, and using a search engine to complete prediction tasks.

[0162] In the example, output module 820 can output the operation information described above.

[0163] The AI ​​agent 800 according to embodiments of this disclosure can simply and effectively improve the level of intelligence, as well as enhance flexibility and versatility.

[0164] Figure 9 A schematic block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0165] like Figure 9 As shown, device 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 902 or a computer program loaded from storage unit 908 into random access memory (RAM) 903. RAM 903 may also store various programs and data required for the operation of device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0166] Multiple components in device 900 are connected to I / O interface 905, including: input unit 906, such as keyboard, mouse, etc.; output unit 907, such as various types of monitors, speakers, etc.; storage unit 908, such as disk, optical disk, etc.; and communication unit 909, such as network card, modem, wireless transceiver, etc. Communication unit 909 allows device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0167] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as the large model-based information update method. For example, in some embodiments, the large model-based information update method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on device 900 via ROM 902 and / or communication unit 909. When the computer program is loaded into RAM 903 and executed by the computing unit 901, one or more steps of the large model-based information update method described above can be performed. Alternatively, in other embodiments, the computing unit 901 can be configured to perform the large model-based information update method by any other suitable means (e.g., by means of firmware).

[0168] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0169] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0170] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0171] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0172] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0173] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0174] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0175] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information update method based on a large model, comprising: A large model is used to analyze historical information associated with the target task to obtain information to be operated on. This historical information is acquired by the agent during the execution of the target task based on its task requirements. The information to be operated on indicates the operation type and target information for performing an update operation on the object to be updated in the historical information. There is a difference between the historical content information of the object to be updated and the input information used to execute the target task. According to the operation type, by performing an update operation on the object to be updated, the historical content information of the object to be updated is updated to the target information.

2. The method according to claim 1, wherein, The process of analyzing historical information associated with the target task using a large model to obtain information to be operated on includes: By analyzing historical information associated with the target task using a large model, candidate operation information is obtained; wherein, the candidate operation information indicates the candidate operation type for performing update operations on each of the multiple candidate objects in the historical information; and The operation information to be performed is determined from the candidate operation information.

3. The method according to claim 2, wherein, The candidate operation information also includes interaction scenario information applicable to each of the multiple candidate objects; Determining the operation information from the candidate operation information includes: From the plurality of candidate objects, determine the object to be updated that matches the historical information; as well as In response to determining that the interaction scenario information applicable to the object to be updated matches the execution environment information of the target task, the operation type associated with updating the object to be updated is determined to be the operation information to be operated.

4. The method according to any one of claims 1-3, wherein, The information to be operated on includes the expected updated information for the object to be updated; the method further includes: In response to determining that the historical content information of the object to be updated and the expected updated information each apply to the same interaction scenario information, semantic analysis is performed on the historical content information of the object to be updated and the expected updated information to obtain semantic similarity; and Based on the semantic similarity, the target information is generated using the large model based on the historical content information and the expected updated information.

5. The method according to claim 4, wherein, The step of generating the target information based on the semantic similarity using the large model based on the historical content information and the expected updated information includes: In response to determining that the semantic similarity is less than a predetermined threshold, the target information is generated by semantically merging the historical content information and the expected updated information using the large model.

6. The method according to claim 4 or 5, wherein, The step of generating the target information based on the semantic similarity using the large model based on the historical content information and the expected updated information includes: In response to determining that the semantic similarity is greater than or equal to a predetermined threshold, semantic analysis is performed on the historical content information and the expected updated information using the large model to obtain analysis results; wherein, the analysis results indicate whether there is a semantic contradiction between the historical content information and the expected updated information; and In response to determining that the analysis result indicates a semantic contradiction between the historical content information and the expected updated information, the expected updated information is determined to be the target information.

7. The method according to any one of claims 1-6, wherein, The step of updating the historical content information of the object to be updated to the target information by performing an update operation on the object to be updated according to the operation type includes: The update task information is determined based on the target information and the operation type information; wherein, the update task information includes the operation information for each of the multiple objects to be updated; and The update task is executed asynchronously with the target task, and update operations are performed on each of the multiple objects to be updated.

8. The method according to any one of claims 1-7, further comprising: Obtain the first historical information associated with the task requirements of the task to be executed from the candidate information; as well as Using an intelligent agent, the task to be executed is performed based on the information obtained by concatenating the first historical information and the information required by the task; wherein, the candidate information includes at least one of the target information.

9. The method according to claim 8, wherein, The step of using an intelligent agent to execute the task to be performed based on information concatenated from the first historical information and the task requirement information includes: The information to be retrieved is generated by the intelligent agent by splicing the first historical information and the information of the task requirements. Obtain second historical information associated with the information to be retrieved from the candidate information; and The intelligent agent executes the task to be performed based on the first historical information, the second historical information, and the task requirements.

10. An information updating device based on a large model, comprising: An analysis module is used to analyze historical information associated with a target task using a large model to obtain information to be operated on. This historical information is acquired by the agent during the execution of the target task based on its task requirements. The information to be operated on indicates the operation type and target information for performing an update operation on an object to be updated within the historical information. There is a difference between the historical content information of the object to be updated and the input information used to execute the target task. The execution module is used to update the historical content information of the object to be updated to the target information by performing an update operation on the object to be updated according to the operation type.

11. An intelligent agent, comprising: An input module is used to receive historical information associated with a target task; wherein the historical information is obtained by another intelligent agent during the execution of the target task based on the task requirements of the target task; The processing module is configured to determine a target task based on the historical information received by the input module, determine a target large model based on the target task, and obtain the operation-to-operate information by invoking the target large model to execute the method described in any one of claims 1-9; the operation-to-operate information indicates the operation type and target information for performing an update operation on the object to be updated in the historical information; there is a difference between the historical content information of the object to be updated and the input information used to perform the target task; and An output module is used to output the operation-to-be-operated information obtained by the processing module, so that, according to the operation type, the historical content information of the object to be updated is updated to the target information by performing an update operation on the object to be updated.

12. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

13. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-9.

14. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-9.

Images