Method and device for processing human-computer interaction information

Abstract

This application relates to the field of data processing technology and provides a method and apparatus for processing human-computer interaction information. The processing method includes, when a user interacts with a device for a target application, storing the generated interaction information in a first storage layer, where the first storage layer stores a preset number of the latest interaction information according to the interaction time; determining summary information and storing it in a second storage layer of a data storage system when a preset summary update condition is met; and storing the summary information stored in the second storage layer and the currently stored interaction information in the first storage layer in a third storage layer of the data storage system when a preset fixed update condition is met. This processing method replaces all historical interaction information with summary information, achieving the storage of long-term context information, thereby saving storage resources required for storing long-term context information and reducing the storage pressure on the device.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and device for processing human-computer interaction information. Background Technology

[0002] With the development of AI (Artificial Intelligence), AI software can be installed in corresponding devices (such as mobile phones and other electronic devices, or vehicles and other transportation equipment) to interact with users. Taking a vehicle as an example, users can interact with the built-in AI agent within the vehicle. The built-in AI agent needs to combine long-term contextual information (needs, habits, historical operations, etc.) generated from the conversation with the user to understand the user's intent and perform processing such as application generation and vehicle control operations.

[0003] In related technologies, the retention of long-term context information is mainly achieved by saving all historical session records. That is, all session messages between the user and the device are stored in the device's local database for the device's built-in AI agent to read and recognize the user's intent.

[0004] While this retention method can preserve all interaction details and long-term context information, device-side storage resources are limited, and complete saving will occupy a large amount of storage space, resulting in excessive storage pressure on the device. Summary of the Invention

[0005] In view of this, this application aims to propose a method for processing human-computer interaction information, so as to save the storage resources occupied by the device and reduce the storage pressure on the device.

[0006] To achieve the above objectives, the technical solution of this application is implemented as follows:

[0007] A method for processing human-computer interaction information, applied to a device with a data storage system, the method comprising: when a user interacts with the device for a target application, storing the generated interaction information in a first storage layer of the data storage system, the first storage layer being used to store a preset number of the latest interaction information according to the interaction time; when a preset summary update condition is met, determining the summary information of the interaction information currently stored in the first storage layer, and storing the summary information in a second storage layer of the data storage system; when a preset fixation update condition is met, storing the summary information stored in the second storage layer and the interaction information currently stored in the first storage layer in a third storage layer of the data storage system.

[0008] Furthermore, the conditions for satisfying the preset summary update conditions include: the interaction information generated by the user interacting with the device contains preset demand adjustment keywords; or, the number of times the summary information stored in the second storage layer has not been updated reaches a preset round threshold; or, a summary information update instruction triggered by the user is received; or, the idle time of the interaction between the user and the device reaches a preset time.

[0009] Furthermore, the conditions for satisfying the preset solidification update conditions include: receiving a save operation instruction triggered by the user for the target application.

[0010] Furthermore, when storing the summary information stored in the second storage layer to the third storage layer, the processing method includes: binding the summary information stored in the second storage layer with the user's user identifier and the application identifier of the target application; and storing the summary information bound with the user identifier and the application identifier to the third storage layer.

[0011] Furthermore, determining the summary information of the interaction information currently stored in the first storage layer includes: extracting historical summary information generated by the user for the target application from the second storage layer; and determining the summary information of the interaction information currently stored in the first storage layer using a preset incremental merging and updating method based on the historical summary information and the interaction information stored in the first storage layer.

[0012] Furthermore, the preset incremental merging and updating method includes: when the interaction information stored in the first storage layer is a preset invalid type of information, determining the historical summary information as the summary information of the interaction information currently stored in the first storage layer; when the interaction information stored in the first storage layer is a preset conflict type of information that conflicts with the historical summary information, correcting the historical summary information according to the interaction information stored in the first storage layer to obtain the summary information of the interaction information currently stored in the first storage layer; when the interaction information stored in the first storage layer is a preset new demand type of information, supplementing the historical summary information according to the interaction information stored in the first storage layer to obtain the summary information of the interaction information currently stored in the first storage layer.

[0013] Furthermore, the processing method further includes: when the preset summary update condition is met, determining the user's current preference information based on the interaction information stored in the first storage layer and the user's historical preference information stored therein; and when the preset fixed update condition is met, storing the user's current preference information in the third storage layer after binding it with at least the user's user identifier.

[0014] Furthermore, when the device responds to the user's trigger and loads historical interaction information for the target application, the processing method further includes: searching for the user's latest historical summary information for the target application in the third storage layer and loading the latest historical summary information into the second storage layer; reading the user's historical interaction information for the target application from the third storage layer and loading it into the first storage layer.

[0015] Furthermore, the processing method further includes: comparing the latest historical summary information with the historical interaction information stored in the third storage layer; if there is no preset correspondence between the latest historical summary information and the historical interaction information, generating historical summary information corresponding to the historical interaction information based on the historical interaction information stored in the third storage layer; and loading the generated historical summary information corresponding to the historical interaction information into the second storage layer.

[0016] Compared with related technologies, this application has at least the following advantages.

[0017] (1) The processing method described in this application compresses and extracts historical interaction information by storing a preset number of the latest interaction information in the first storage layer according to the interaction time and storing summary information in the second storage layer; and persistently stores the summary information in the second storage layer and the interaction information in the first storage layer when the preset persistence update conditions are met. In this way, using summary information to replace the interaction information of historical sessions for storage can save the local storage resources of the device and retain the long-term context of the user's interaction with the device for the target application.

[0018] In other words, this application only needs to store summary information (extraction of historical interaction information) and a preset number of the latest interaction information, without storing all historical interaction information, thereby reducing the occupation of device storage resources and reducing the storage pressure on the device.

[0019] (2) The processing method of this application can also trigger the update of summary information when the interaction information generated by the user and the device contains preset demand adjustment keywords. This can capture the user's demand changes for the target application in a timely manner and ensure that the summary information is updated synchronously to reflect the latest demand. Alternatively, the summary information can be updated when the number of times the summary information has not been updated reaches a preset round threshold. This can avoid the situation where the summary information lags behind the interaction information due to long-term lack of updates and ensure the timeliness of the summary information. Furthermore, the summary information can also be updated when the idle time between the user and the device reaches a first preset time. This can also avoid the problem of loss of interaction information due to session interruption. In addition, the summary information can also be updated in response to the summary information update command triggered by the user, thereby meeting the user's need to actively update the summary and improving the user's autonomy in interaction.

[0020] (3) The processing method of this application also solidifies the user-confirmed summary information and recent interaction information to the third storage layer in a timely manner when the user's save operation instruction for the target application is received. This ensures that the user-confirmed summary information will not be lost, thus facilitating subsequent reuse.

[0021] (4) The processing method of this application also binds the summary information with the user identifier and the application identifier and stores it in a third storage layer. This enables the classified storage of summary information. This bound storage method can avoid the confusion of summary information of different users and different target applications, and facilitates the quick retrieval of summary information of a specific user for a specific target application, thereby improving the management efficiency and retrieval accuracy of summary information.

[0022] (5) The processing method of this application further includes, when determining the summary information of the interaction information in the first storage layer, first extracting the user's historical summary information for the target application from the second storage layer, and then combining it with a preset number of the latest interaction information stored in the first storage layer, generating new summary information using a preset incremental merging and updating method. This eliminates the need to re-parse all historical interaction content each time, and only updates the summary information based on the historical summary information and the current latest interaction information, thereby reducing the difficulty and computational load of generating summary information and saving the computing resources of the device; at the same time, updating in conjunction with historical summary information can also ensure the continuity of summary information.

[0023] (6) In the processing method of this application, when determining the summary information, if the interaction information of the first storage layer is a preset invalid type of information, the historical summary information is directly used. This avoids redundancy in summary updates caused by invalid interaction information and saves computing resources. When the interaction information is a preset conflict type of information that conflicts with the historical summary information, the historical summary information is corrected based on the current interaction information to ensure that the summary information can reflect changes in user needs in a timely manner and avoid the situation where the device misunderstands the user's intention due to summary lag. When the interaction information is a preset new requirement type of information, new content is added on the basis of the historical summary, thereby improving the summary information and thus improving the accuracy of the device in understanding the user's intention.

[0024] (7) The processing method of this application also determines the user's current preference information by combining the interaction information of the first storage layer and the user's historical preference information when the preset summary update conditions are met. In this way, it is possible to capture the user's habits and tendencies in the interaction process of the target application; when the preset solidification update conditions are met, the current preference information is bound to the user identifier and stored in the third storage layer, which can realize the long-term retention of the user's preference information. Subsequently, the user does not need to repeat their own preferences, and the preference information can be directly called to provide personalized interaction services to the user, thereby improving the personalization of human-computer interaction and user experience.

[0025] (8) The processing method of this application further includes, when the device responds to a user trigger to load the historical interaction information of the target application, searching for and loading the latest historical summary information of the user for the target application from the third storage layer to the second storage layer, and simultaneously reading the corresponding historical interaction information to the first storage layer, thereby restoring the interaction information generated in previous sessions. In this way, when the user interacts with the device again for the target application, the device can directly understand the user's intent based on the loaded summary information and historical interaction information, without requiring the user to repeat the description of historical interaction content, thereby improving the continuity of interaction.

[0026] (9) The processing method of this application also compares the latest historical summary information with the historical interaction information of the third storage layer when loading historical interaction information; when the two are not in a preset correspondence, the corresponding historical summary information is regenerated based on the historical interaction information and loaded into the second storage layer to ensure that the summary information loaded into the second storage layer can completely and accurately reflect the historical interaction content, and avoid the situation where the device misunderstands the user's historical needs due to the mismatch between the summary information and the historical interaction information.

[0027] Another objective of this application is to provide a device comprising a memory and a processor; the memory having a data storage system including a first storage layer, a second storage layer and a third storage layer, and storing a computer program in the memory; when the processor executes the computer program, it implements the aforementioned method for processing human-computer interaction information.

[0028] The device described in this application implements the above-mentioned human-computer interaction information processing method, and replaces all historical interaction information with summary information to realize the storage of long-term context information, thereby saving the storage resources required to store long-term context information. Attached Figure Description

[0029] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0030] Figure 1 This is a flowchart illustrating the human-computer interaction information processing method described in an embodiment of this application.

[0031] Figure 2 This is a flowchart illustrating the process of determining summary information in the human-computer interaction information processing method described in this application embodiment.

[0032] Figure 3 This is a schematic diagram illustrating the implementation process of the preset incremental merging and updating method in the human-computer interaction information processing method described in the embodiments of this application.

[0033] Figure 4 This is a schematic diagram of the process for determining and storing user preference information in the human-computer interaction information processing method described in the embodiments of this application.

[0034] Figure 5 This is a schematic diagram illustrating the complete process of generating new interactive information in the human-computer interaction information processing method described in the embodiments of this application.

[0035] Figure 6 This is a schematic diagram illustrating the complete process of loading historical sessions in the human-computer interaction information processing method described in this application embodiment. Detailed Implementation

[0036] To make the technical solution and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0038] Furthermore, in the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0039] In this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0040] The present application will now be described in detail through exemplary embodiments. However, it should be understood that, without further description, elements, structures, and features in one embodiment may be advantageously incorporated into other embodiments.

[0041] An embodiment of the first aspect of this application provides a method for processing human-computer interaction information, which is applied to a device with a data storage system. The device includes, but is not limited to, electronic devices with human-computer interaction and AI interaction functions such as smartphones, tablets, smart wearable devices, smart speakers, and smart home devices, and may also include transportation equipment such as vehicles.

[0042] With the development of AI (Artificial Intelligence), users can interact with AI agents built into devices through conversations.

[0043] In this embodiment, the device is taken as a transportation device, specifically a vehicle, for illustration. The vehicle's in-vehicle infotainment system is equipped with an in-vehicle AI agent, which users can interact with within the vehicle. The in-vehicle AI agent needs to understand the user's intent and perform corresponding processing based on that intent, such as generating in-vehicle applications for the user, performing vehicle control operations, and providing personalized services.

[0044] For example, users can interact with the in-vehicle AI assistant to generate a lighting control app (application) within the vehicle's infotainment system, which integrates control of headlights such as low beam, high beam, and fog lights. Users can then directly control the vehicle's lights through this app.

[0045] For in-vehicle AI agents, it is necessary to understand user intent based on long-term contextual information such as user needs, habits, and historical operations generated from past user conversations. This long-term contextual information is crucial for in-vehicle AI agents to accurately understand user intent; therefore, it is necessary to store and retain this long-term contextual information within the vehicle.

[0046] In related technologies, there are two main ways to preserve long-term context information.

[0047] One approach is to fully preserve all historical session records, such as storing all conversation messages between the user and the in-vehicle AI agent in the vehicle's local database. This allows the in-vehicle AI agent to read all conversation messages and identify the user's intent. While this method preserves all interaction details, the limited local storage resources in the vehicle mean that full storage would consume a significant amount of storage space.

[0048] To overcome the storage pressure caused by saving the entire conversation, another approach is to save only a small number of recent interactions. While this method saves storage space and improves response speed, it loses a significant amount of historical interaction content, making it prone to errors when the in-vehicle AI agent understands the user's intent.

[0049] It is evident that related technologies cannot strike a balance between preserving details and consuming storage resources when storing long-term contextual information. If all historical dialogue records are fully saved, it will lead to storage resource shortages. If only a small number of recent dialogues are saved, it will result in the loss of a large amount of long-term contextual information, which in turn will affect the accuracy of the device's built-in AI agent in understanding the user's intent.

[0050] In view of this, in order to overcome the shortcomings of related technologies, the human-computer interaction information processing method in this embodiment combines... Figure 1 In terms of overall design, it includes the following steps S110-S130.

[0051] Step S110: When the user and the device interact with the target application, the generated interaction information is stored in the first storage layer of the data storage system.

[0052] Specifically, user interaction with a device can be, for example, interaction between the user and an AI agent within the device. This AI agent refers to AI software; for instance, in a vehicle, this AI agent could be an in-vehicle AI assistant or an AI application generator.

[0053] A target application refers to an application created, configured, or used by a user through interaction with an AI agent. For example, in a scenario involving interaction with a vehicle, the target application might be a lighting control application, an air conditioning application, an entertainment application, etc.

[0054] For example, user Zhang San has a conversation with the in-vehicle AI agent inside the car. The conversation content is as follows.

[0055] First round of conversation: "User: "Give me a headlight control application." In-vehicle AI agent: "Hello! Regarding headlight control, I can generate different applications for you. You can choose the following functions as needed: (1) Headlight control: switch between low beam, high beam, and automatic mode; (2) Rear fog light: turn the rear fog light on / off; (3) Reading light: control the interior reading light; (4) Combined application: such as "headlight + rear fog light", "reading light + rear fog light", etc. Which one would you like to generate? Tell me your choice."

[0056] Second round of conversation: "User: I want them all. In-vehicle AI assistant: Great! I've generated a 'Full Vehicle Lighting Control' app for you, which now integrates the following functions: Headlight control—supports low beam, high beam, and automatic mode switching; Rear fog lights—enhances visibility in inclement weather; Reading lights—facilitate nighttime reading for passengers. If you'd like to create a more immersive in-vehicle lighting atmosphere, I can also add: ambient light switch, lighting effect modes (supports music rhythm); let me know your ideas anytime!"

[0057] Third round of conversation: "User: Added lighting effect modes; In-vehicle AI intelligent agent: We have successfully upgraded your 'Full Vehicle Lighting Control' application! Now added: Ambient light switch - the key to turning on the lighting atmosphere; supports music rhythm, breathing, multi-color and other dazzling effects; now you can not only control functional lights, but also create your own exclusive in-vehicle lighting atmosphere~ If you also want the lights to dance with the music, I can also recommend entertainment sound effect linkage settings for you, let me know anytime!"

[0058] In other words, when user Zhang San converses with the in-vehicle AI agent inside the car, requesting the in-vehicle AI agent to generate a corresponding headlight control application, the user is Zhang San, and the target application is equivalent to the headlight control application.

[0059] This interactive information refers to the conversational messages generated by the user's interaction with the AI ​​agent. More specifically, each round of conversation between the user and the device's built-in AI agent is equivalent to generating one piece of interactive information. For example, if the user issues a command and the in-vehicle AI agent provides a response, the command and response together constitute one round of conversation, which is equivalent to one piece of interactive information. As another example, the interaction in the above example generated three rounds of conversational messages, which is equivalent to generating three pieces of interactive information.

[0060] In order to save storage resources and ensure the accuracy of user intent recognition of AI agents, in this embodiment, the device's data storage system is divided into a first storage layer, a second storage layer, and a third storage layer.

[0061] The first storage layer is a short-term session storage layer deployed in memory. It stores a preset number of the latest interaction information based on the interaction time. The interaction time refers to the time when the interaction information is generated. The preset number can be set by the user according to the actual situation. For example, it can be set to 8 rounds. In this way, the first storage layer can only retain the interaction information of the most recent 8 rounds. Assuming that the interaction information of rounds 1 to 8 is currently stored, when the interaction information of round 9 is generated, the interaction information of round 1 (the earliest generated interaction information) will be deleted, so that the first storage layer stores the interaction information of rounds 2 to 9.

[0062] The second storage layer is a dynamic summary storage layer, deployed in memory, used to store summary information generated based on recent interactions. The third storage layer is a persistent (or hardened) storage layer, which can be implemented through a local database, used to store relevant information for a long time.

[0063] That is, in step S110, each time the user has a conversation with the AI ​​agent, a round of interaction information will be generated. This round of interaction information is first stored in the first storage layer. If the number of interaction information stored in the first storage layer exceeds the preset number, the interaction information with the earliest generation time stored in the first storage layer is deleted, and the latest generated interaction information is stored in the first storage layer, so that the first storage layer always stores only the preset number of the latest interaction information.

[0064] Step S120: If the preset summary update conditions are met, determine the summary information of the currently stored interactive information in the first storage layer, and store the summary information in the second storage layer of the data storage system.

[0065] Specifically, summary information refers to the core information obtained by understanding, summarizing and compressing information from multiple rounds of interaction. It can reflect the user's core needs, functional configurations, modification history and key operations for the target application.

[0066] Specifically, the summary information should include: 1. Core user needs / goals, such as: in scenarios involving interaction with the in-vehicle AI agent, these core user needs / goals indicate which vehicle control functions the user wants to control, what types of content services (news, music, weather, etc.) they want to subscribe to / browse, and their desired driving / travel experience. 2. Key modification history (briefly linked in chronological order), including records of major changes related to the target application (e.g., adding / deleting / replacing application functions, modules, combinations, etc.). 3. Effective responses / key actions completed by the AI ​​agent in the device. For example: which applications have been created for the user (including which functions / components), what naming adjustments have been made, and which parameters or modules have been modified. Additionally, if there are application designs or function combinations that the user has explicitly rejected, a brief explanation can be provided to avoid repeatedly recommending the same solution later.

[0067] In addition, summary information is usually not static. As users interact with the AI ​​agent built into the device, user needs may be added, modified, or canceled. Therefore, when the preset summary update conditions are met, it is often necessary to determine the summary information and store it in the second storage layer so that the summary information can accurately reflect the latest user intent.

[0068] When storing summary information in the second storage layer of the data storage system, an update storage method is used. Update storage means that the newly generated summary information overwrites or merges the existing old summary information in the second storage layer. For the same user and the same application, multiple copies of the summary information will not be stored repeatedly.

[0069] For example, in the example of step S110 above, after the user completes three rounds of conversation with the in-vehicle AI agent, if a fourth round of conversation occurs, for example, the fourth round of conversation is: "User: Change to an elegant name. In-vehicle AI agent: I have renamed the application to Light Domain (such as the Realm of Light and Shadow, all under control). If you like this style, you can also consider these elegant names: Smart Light, Star Language, Flowing Light. Tell me your favorite name anytime, and I will change it for you immediately~".

[0070] Furthermore, if the preset summary update conditions are met at this time, the first storage layer stores the four rounds of interaction information. Based on these four rounds of interaction information, corresponding summary information is generated. The generated summary information could be, for example, "The user wants to create an in-vehicle application that fully controls all vehicle lights. The initial requirements include headlights, rear fog lights, and reading lights, which were later upgraded to add ambient light switches and lighting effect modes (supporting music rhythms, etc.). The in-vehicle AI agent has created and continuously optimized this application, which was finally named 'Light Domain' to achieve integrated control of functionality and ambient lighting."

[0071] Step S130: If the preset solidification and update conditions are met, store the summary information stored in the second storage layer and the interaction information currently stored in the first storage layer into the third storage layer of the data storage system.

[0072] Specifically, when a user confirms a request, saves the application, or the session remains idle for an extended period, the dynamic summary in memory, along with recent interaction information, needs to be persistently stored for reuse across sessions. Therefore, upon detecting that a preset persistence update condition is met, the summary information stored in the second storage layer is used as the persistence summary and stored in the third storage layer, and persistently stored in the local database. A preset number of interaction messages currently stored in the first storage layer are also persistently stored in the third storage layer.

[0073] Through steps S110-S130, a preset number of the latest interaction information are stored in the first storage layer according to the interaction time, and summary information is stored in the second storage layer, thus compressing and refining historical interaction information. Furthermore, when preset persistence and update conditions are met, the summary information from the second storage layer and the interaction information from the first storage layer are persistently stored. In this way, using summary information to replace historical session interaction information for storage saves local storage resources on the device while preserving the long-term context of user interactions with the device for the target application.

[0074] In other words, this embodiment only needs to store summary information (extraction of historical interaction information) and a preset number of the latest interaction information. It does not need to store all historical interaction information, thereby reducing the local storage resource occupation of the device and reducing the storage pressure on the device.

[0075] Furthermore, by using summary information to retain long-term contextual information, the device's AI agent can accurately understand user intent and improve the user experience.

[0076] It is worth noting that after the current interaction information is generated, if the preset persistence update conditions are met, the interaction information in the first storage layer will be stored in the third storage layer for persistent storage. If new interaction information is subsequently generated for the session of this target application, and persistent storage is triggered again, in some embodiments, the latest interaction information currently stored in the first storage layer can overwrite the older persistent interaction information. In this way, the third storage layer only retains a preset number of the latest interaction information corresponding to the target application.

[0077] Continue to combine Figure 1 As shown, in some exemplary implementations, the conditions for satisfying the preset summary update conditions in step S120 above may specifically include: at least one of the following four update sub-conditions is satisfied.

[0078] The first update sub-condition is: the interaction information generated by the user's interaction with the device contains preset adjustment keywords.

[0079] For example, after a user has already generated four rounds of interaction information with the vehicle's in-vehicle AI agent, a fifth round of interaction information is generated. If the fifth round of interaction information contains preset demand adjustment keywords, it indicates that the first update sub-condition is met.

[0080] The preset demand adjustment keywords are words indicating changes in user needs, such as additions, modifications, deletions, or replacements. Examples include "add," "modify," "delete," "don't want," and "replace." When preset demand adjustment keywords are present in the interaction information, it usually indicates a substantial change in the user's needs regarding the target application. If the summary information is not updated at this point, it can easily lead to inconsistencies between the summary information and the user's true intent, thus affecting the accuracy of the AI ​​agent's understanding of the user's needs. Therefore, when preset demand adjustment keywords are detected in the interaction information, the preset summary update conditions are determined to be met, triggering the summary information update operation.

[0081] More specifically, interactive information can be identified through text matching, semantic recognition, and other means to determine whether the preset requirement adjustment keyword exists; if at least one preset requirement adjustment keyword exists, the first update sub-condition is determined to be met.

[0082] The second update sub-condition is: the number of times the summary information stored in the second storage layer has not been updated reaches a preset round threshold.

[0083] Specifically, the number of times the summary information has not been updated refers to the number of new conversational interaction rounds between the user and the AI ​​agent since the last generation or update of the summary information. For example, after the summary information was generated, there were 3 more conversations and 3 new interaction information were obtained. Now, a 4th conversation has been generated, and the number of times the summary information has not been updated is 4.

[0084] The preset round threshold is a pre-set update interval number of rounds, which can be set by the user according to actual needs, such as 4 rounds. It is worth noting that in this embodiment, the preset round threshold should not exceed the preset amount of interactive information that the first storage layer can store. This ensures that all newly generated interactive information after the last generation or update of summary information is retained in the first storage layer, without any new interactive information being deleted due to the first storage layer exceeding its storage limit. When updating the summary information, all newly generated interactive information can be used for updating, ensuring that the generated summary information is complete and accurate, avoiding incomplete summary information due to the deletion of some information.

[0085] When the number of times the summary information has not been updated is greater than or equal to the preset round threshold, it indicates that enough new interaction information has been accumulated recently, and a summary update operation needs to be triggered to update the summary information and store the new summary information update in the second storage layer.

[0086] The third update sub-condition is: receiving a summary information update instruction triggered by the user.

[0087] The summary information update command is a user-initiated command used to force an update of the summary. For example, during normal interaction between the user and the in-vehicle AI agent, if the user acknowledges the response of the in-vehicle AI agent, or believes that the planned application creation, update, or modification has been completed, the user may reply with "Confirm Request" or "Save" in the dialogue, or the user may click "Confirm" or "Save" buttons on the vehicle's interface, all of which are considered to trigger the summary information update command.

[0088] For example, after the in-vehicle AI agent creates a headlight control application for the user, the user needs to save the application before it can be generated for their use. During the dialogue between the user and the in-vehicle AI agent, if the user believes they have completed the application creation process, they can click the relevant save button to complete the application creation. Clicking the save button at this point indicates that the user has triggered a summary information update command, requiring an update of the summary information.

[0089] The fourth update sub-condition is: the idle time of interaction between the user and the device reaches the preset time.

[0090] Idle time refers to the period of no interaction from the moment the last conversation message (the last interactive information) is generated until the current moment. For example, if a user completes one round of interaction with the in-vehicle AI agent and there is no new interaction for 10 consecutive minutes, then the idle time is 10 minutes.

[0091] This preset duration can be set by the user according to their in-vehicle interaction habits, for example, 8-12 minutes. When the idle duration is greater than or equal to this preset duration, it indicates that the current session may be temporarily terminated. To ensure the completeness and accuracy of the summary information, an update needs to be triggered to avoid loss of summary information due to session interruption. When the idle duration is less than the preset duration, it indicates that the fourth update sub-condition is not met.

[0092] Therefore, by using these preset summary update conditions, summaries can be updated when user-device interaction information contains preset adjustment keywords. This allows for timely capture of changes in user needs for the target application, ensuring that summary information is updated synchronously to reflect the latest requirements. Alternatively, summaries can be updated when the number of times they haven't been updated reaches a preset threshold, preventing them from lagging behind interaction information due to prolonged inactivity and ensuring timeliness. Furthermore, summaries can be updated when the idle time between the user and device reaches a first preset duration, preventing information loss due to session interruptions. Additionally, summaries can be updated in response to user-triggered update commands, satisfying users' need to proactively update summaries and enhancing user autonomy.

[0093] In other words, by using multi-dimensional summary update triggering conditions, the timeliness of summary updates can be achieved, ensuring that the summary information accurately reflects the user's interaction intent.

[0094] Continue to combine Figure 1 As shown, in some exemplary implementations, the conditions for satisfying the preset firmware update conditions in step S130 may specifically include: receiving a save operation instruction triggered by the user for the target application.

[0095] The save operation instruction refers to an operation instruction initiated by the user to confirm and retain the relevant interactive content of the target application for a long time. It can be either an interface operation instruction or a voice instruction.

[0096] For example, after the user completes the application creation or configuration, or after the user confirms that the current requirements are met, clicking the "Save Application" or "Confirm Generation" buttons on the vehicle's infotainment system will trigger a save operation command.

[0097] Additionally, when a user confirms their current interaction needs via voice or other means (such as when the user says "confirm current settings"), it can also be considered as receiving a save operation command triggered by the user for the target application.

[0098] Upon receiving a save command, the system automatically saves the summary information and the short-term interaction information stored in the first storage layer to the third storage layer to prevent data loss.

[0099] Therefore, when a user sends a save command for a target application, the user-confirmed summary information and recent interaction information are promptly stored in the third storage layer. This ensures that the user-confirmed summary information is not lost, thus facilitating subsequent reuse.

[0100] Furthermore, in some embodiments, when the summary information is triggered to be updated and stored in the second storage layer in step S120 above, if the current summary information update is caused by detecting that the idle time of the interaction between the user and the device has reached a preset time, or by receiving a summary information update instruction triggered by the user, the summary information stored in the second storage layer can be used as the solidified summary information and temporarily stored in the third storage layer. After detecting that the preset solidification update conditions are met, the temporarily stored solidified summary information and the interaction information stored in the first storage layer are then persistently stored in the local database.

[0101] Continue by Figure 1 and combined Figure 2 As shown, in some exemplary embodiments, the step S120 above, which determines the summary information of the currently stored interactive information in the first storage layer, may specifically include the following steps S121-S122.

[0102] Step S121: Extract historical summary information generated by the user for the target application from the second storage layer.

[0103] For example, the latest historical summary information is originally stored in the second storage layer. When updating the summary information this time, the historical summary information generated by the user for the target application is first read from the second storage layer.

[0104] Step S122: Based on the historical summary information and the interaction information stored in the first storage layer, determine the summary information of the interaction information currently stored in the first storage layer using a preset incremental merging and updating method.

[0105] Specifically, the historical summary information represents a summary of previous rounds of interaction information, and the interaction information currently stored in the first storage layer is only a supplement or correction based on the historical information. Therefore, in step S122, the historical summary information is incrementally merged or updated based on the interaction information currently stored in the first storage layer to determine the latest summary information.

[0106] Therefore, through steps S121-S122, when determining the summary information of the interaction information in the first storage layer, it is necessary to combine the summary information obtained by analyzing the historical long-term context and the current short-term session, and store it in the second storage layer. Thus, in this embodiment, the historical summary information of the user for the target application is first extracted from the second storage layer. This historical summary information is equivalent to the historical long-term context. Then, combined with a preset number of the latest interaction information stored in the first storage layer (equivalent to the current short-term session), a preset incremental merging update method is used to generate new summary information. This eliminates the need to re-parse all historical interaction content each time; the summary information is updated only based on the historical summary information and the current latest interaction information. This reduces the difficulty and computational load of generating summary information, thereby saving device computing resources. Simultaneously, updating in conjunction with historical summary information also ensures the continuity of the summary information.

[0107] Specifically, refer to Figures 1 to 3 In some exemplary implementations, the preset incremental merge update method in step S122 may specifically include the following steps S1221-S1223.

[0108] Step S1221: If the interaction information stored in the first storage layer is a preset invalid type of information, the historical summary information is determined as the summary information of the interaction information currently stored in the first storage layer.

[0109] The default invalid message type refers to interaction information that does not involve the target application. More specifically, the default invalid message type is a message that is unrelated to the target application's functional requirements, operation adjustments, parameter changes, interaction logic, etc., that is, it does not contain any information that has a substantial impact on the target application. Examples include polite language, small talk, and meaningless chatter.

[0110] If all the preset number of interaction messages stored in the first storage layer are of the preset invalid type, it means that recent interactions have not had any substantial impact on the target application. Therefore, the historical summary information can be directly used as the latest summary information. That is, if the latest conversation is just a simple confirmation, thank you, or casual chat, without introducing any new facts or key changes, the original historical summary information can be basically used, with only minor modifications or no changes made when necessary.

[0111] Step S1222: If the interaction information stored in the first storage layer is a preset conflict type information that conflicts with the historical summary information, the historical summary information is corrected according to the interaction information stored in the first storage layer to obtain the summary information of the interaction information currently stored in the first storage layer.

[0112] Among them, the preset conflict type information refers to the interactive information that conflicts with the historical summary information. That is, interactive information that is contrary to or contradicts the needs expressed by the historical summary information.

[0113] When the preset number of interaction messages stored in the first storage layer belong to the preset conflict type, it means that the user has clearly modified or denied the previous request. Therefore, the historical summary information needs to be corrected (e.g., deleting old content, adding new content, or overwriting the old request with the new request). The corrected summary information is the summary information of the currently stored interaction information.

[0114] In other words, if the user clearly overturns or modifies the original conclusion in the latest conversation, the summary information should be updated with the information corresponding to the latest conclusion; and the evolution from the original solution to the new solution can also be briefly explained, but it is not necessary to describe all the details of the conversation.

[0115] Step S1223: If the interaction information stored in the first storage layer is a preset new requirement type information, supplement the historical summary information according to the interaction information stored in the first storage layer to obtain the summary information of the interaction information currently stored in the first storage layer.

[0116] Among them, the preset new requirement type information refers to the interactive information that includes users' new requirements, new functions, and new configurations for the target application.

[0117] If the interaction information stored in the first storage layer is a preset new demand type, it means that the user has added new content based on the original demand. Therefore, it is necessary to supplement the historical summary information. The supplemented summary information is the summary information of the currently stored interaction information.

[0118] Specifically, supplementing historical summary information can include adding new features, configurations, constraints, or other new content to the existing summary while maintaining the original summary structure and only expanding the content.

[0119] Therefore, through steps S1221-S1223, when determining the summary information, if the interaction information in the first storage layer is a preset invalid type of information, the historical summary information is directly used. This avoids redundancy in summary updates caused by invalid interaction information and saves computing resources. When the interaction information is a preset conflict type of information that conflicts with the historical summary information, the historical summary information is corrected based on the current interaction information to ensure that the summary information can reflect changes in user needs in a timely manner, avoiding situations where the device misunderstands the user's intent due to summary lag. When the interaction information is a preset new requirement type of information, new content is added to the historical summary, thereby improving the summary information and enhancing the accuracy of the device's understanding of the user's intent. This setting ensures both the accuracy and consistency of the summary information and improves the efficiency of summary information updates.

[0120] Continue to combine Figure 1 As shown, in some exemplary embodiments, step S130 above, when storing the summary information stored in the second storage layer to the third storage layer, may specifically include:

[0121] The summary information stored in the second storage layer is bound to the user's user identifier and the target application's application identifier; the summary information bound to the user identifier and application identifier is then stored in the third storage layer.

[0122] Specifically, user identifiers refer to information used to uniquely identify a user, such as user ID (identifier) ​​or account ID. Application identifiers refer to the identifiers of a target application, such as application ID or application number.

[0123] In addition, in some embodiments, agent identifiers can be bound, with different in-vehicle AI agents corresponding to different agent identifiers. This can distinguish different in-vehicle AI agents and achieve information isolation among multiple agents.

[0124] In this way, summary information is bound to user and application identifiers and then permanently stored in the third storage layer. This allows summary information to be categorized and stored by user and application. This bound storage method avoids confusion between summary information from different users and different target applications, and facilitates quick retrieval of summary information for a specific user and a specific target application, thereby improving the management efficiency and retrieval accuracy of summary information.

[0125] Continue to combine Figure 1 and Figure 4 As shown, in some exemplary embodiments, the processing method may further include the following steps S410-S420.

[0126] Step S410: If the preset summary update conditions are met, determine the user's current preference information based on the interaction information stored in the first storage layer and the stored user's historical preference information.

[0127] Step S420: If the preset fixed update conditions are met, the user's current preference information is bound to the user's user identifier and then stored in the third storage layer.

[0128] Specifically, preference information refers to information extracted from user interactions, such as habits, tendencies, and frequently used operation combinations. Examples include users' preferred vehicle control functions, the order in which functions are used, and scenario-based configurations. For instance, in the example of step S120 above, the determined preference information could be: "Users tend to gradually add new functions based on existing applications, prefer to combine and integrate vehicle control components (such as lights); they have certain aesthetic requirements for application naming, accepting names with poetic and atmospheric qualities (such as light fields); in function selection, they consider both practicality and experience, and are willing to expand to enhanced modules such as ambient lighting effects."

[0129] More specifically, this preference information may include application function and module preferences, function priority / usage order, function combination and dependency preferences, parameter and configuration habits, interaction and AI response habits, naming and application positioning preferences, operation habits and multi-round workflow preferences, as well as other preferences that are strongly related to the generation of in-vehicle applications and can be reused in the long term.

[0130] Application function and module preferences can be categorized as follows: which type of in-vehicle application functions or modules users prefer (e.g., vehicle control, information, entertainment, travel services, etc.); and which modules they prioritize when selecting functions (e.g., whether they are more concerned with vehicle control and safety-related functions or content ecosystem / entertainment-related functions).

[0131] Function priority / usage order indicates the core functions that users prioritize, as well as secondary but optional functions; for example, "prioritize basic vehicle control functions such as headlights / air conditioning, and then consider ambient lighting and entertainment functions."

[0132] Feature combination and dependency preferences indicate whether users tend to combine multiple types of applications together (e.g., "reading light + rear fog light", "ambient light switch + ambient light mode"), and whether they are sensitive to feature dependencies (e.g., whether they emphasize that dependencies must be included in the generation).

[0133] Parameters and configuration habits indicate whether users tend to provide multiple parameter values ​​in a single request (such as multiple selections of news topics) or add them gradually; and whether they have preferred default parameters or typical configurations (such as frequently viewed news topics, commonly used vehicle control modes, etc.).

[0134] Interaction and AI response habits include preferences for answer style, presentation format, and language / tone. For example, answer style preferences might include whether users prefer directly providing application configuration results or first offering recommended options for the user to choose from. Presentation format preferences might include clear, point-by-point explanations, recommendations organized by scenario, and getting straight to the point. Language / tone preferences might include formal / colloquial language, and whether users prefer more human-like / emotional expressions.

[0135] Naming and application positioning preferences include application name style preferences (such as a preference for short, straightforward, and functional names, or a preference for creative / atmospheric names) and whether there are stable preferences or conventional classifications for module affiliation (such as vehicle control, ecosystem, etc.).

[0136] User habits and multi-round workflow preferences include, for example, whether users prefer to filter through a list of recommended features before confirming and generating the application, or whether they prefer to gradually add / fine-tune features to an existing application or create a completely new application. Also, whether users frequently expand the feature set based on the previous suggestions from the in-vehicle AI assistant.

[0137] Other preferences that are strongly related to in-vehicle application generation and can be reused over a long period of time refer to preference information that may continue to be applicable in multiple future in-vehicle application generation dialogues.

[0138] It is worth noting that if the latest interaction information does not generate new information in a certain preference dimension, then the description of the historical preference information can be retained for that preference dimension.

[0139] In addition, when merging historical preference information with current user preference information to obtain the latest user preferences, the preference merging rules include: only updating dimensions with new preference information, not updating when preferences cannot be extracted from the latest interaction information, and merging if there is no relationship and there is a conflict that needs to be judged.

[0140] For example, if the latest interaction information provides a more specific and up-to-date preference description for a certain dimension, the new information can replace or supplement the description of that dimension in the historical preference information; if the latest interaction information does not contain a new preference description, the historical preference information is kept unchanged and used as the new preference information.

[0141] If the latest session is just a one-off task, a temporary need, or a session that doesn't reveal any long-term preferences, then the user's preference information can remain unchanged to avoid mistaking occasional requests unrelated to long-term habits for user preferences.

[0142] If the preferences determined based on the latest interaction information do not conflict with historical preferences, but are simply new scenarios or habits, they can be merged to obtain the final new preference information.

[0143] If a clear conflict exists, it's necessary to judge based on the context: If the user expresses a tendency to "prefer the new way in the future," then the preference determined by the latest interaction information should be the primary factor for merging and updating. If this clear conflict is merely a new attempt by the user, then it's not necessary to completely cover the history; instead, the description can be marked with "Sometimes I try..., but I still mainly prefer..." to obtain new preference information.

[0144] After determining the user's current preference information, if the preset conditions for fixing and updating are met, step S420 is executed to bind the user's current preference information with the user identifier and store it in the third storage layer. Alternatively, the current preference information can also be bound with the application identifier of the target application and the agent identifier of the in-vehicle AI agent before being stored in the third storage layer.

[0145] It is worth noting that if the preset summary update conditions are not met, step S410 and step S120 are not executed; only step S110 is executed repeatedly. If the preset solidification update conditions are not met, step S120 and step S130 are not executed; only steps S110-S120 are executed repeatedly. These will not be elaborated further here.

[0146] Therefore, through steps S410 and S420, when the preset summary update conditions are met, the user's current preference information is determined by combining the interaction information from the first storage layer and the user's historical preference information. This allows for accurate capture of the user's habits and tendencies during interaction with the target application. When the preset fixation update conditions are met, the current preference information is bound to the user identifier and stored in the third storage layer, enabling long-term retention of the user's preference information. This preference information can then be directly accessed to provide personalized interactive services to the user, eliminating the need for the user to repeatedly state their preferences, thus improving the personalization of human-computer interaction and the user experience.

[0147] At the same time, the method of binding and storing preference information with user identifiers can also avoid confusion between different user preferences.

[0148] Continue to combine Figure 1 As shown, in some exemplary embodiments, when a user interacts with the in-vehicle AI agent and new interaction information is generated, the storage of new interaction information is achieved through the above embodiments. Furthermore, when the in-vehicle AI agent interacts with the user, and the device responds to user triggers by loading historical interaction information for the target application, the processing method may further include:

[0149] The system retrieves the latest historical summary information of the user regarding the target application from the third storage layer and loads this information into the second storage layer. It then reads the user's historical interaction information regarding the target application from the third storage layer and loads it into the first storage layer.

[0150] The latest historical summary information refers to the summary information that has been stored in the third storage layer and generated most recently during the user's previous interactions with the in-vehicle AI agent for the target application. It is used to reflect the user's latest core needs for the target application.

[0151] Historical interaction information refers to a preset amount of interaction information that has been stored in the third storage layer when the user previously interacted with the in-vehicle AI agent for the target application.

[0152] For example, when a user exits a conversation with the in-vehicle AI agent regarding a target application (e.g., closes the application or exits the dialogue interface) and then reopens the conversation, the vehicle will automatically read the latest historical summary information and historical interaction information corresponding to the target application from the third storage layer, restore the latest historical summary information to the second storage layer, so that the second storage layer has a long-term context with the user regarding the target application, and restore the historical interaction information to the first storage layer, so that the first storage layer can restore the details of recent multi-round interactions for the in-vehicle AI agent to use quickly.

[0153] Since the first and second storage layers are storage areas that the in-vehicle AI agent can directly access and call in real time, while the third storage layer is a persistent database, the in-vehicle AI agent cannot quickly retrieve historical summary information and recent interaction information from the third storage layer. Therefore, it is necessary to restore the historical summary information and historical interaction information to the second and first storage layers respectively. This allows the in-vehicle AI agent to directly and quickly call the latest loaded historical summary information and historical interaction information, quickly connecting the previous dialogue context without requiring the user to repeatedly describe clearly defined needs, configurations, etc., thereby improving the continuity of interaction.

[0154] Therefore, when the device responds to a user's trigger to load historical interaction information for the target application, it retrieves and loads the latest historical summary information for that user's interaction with the target application from the third storage layer to the second storage layer, while simultaneously reading the corresponding historical interaction information to the first storage layer, thus restoring previous session information. This way, when the user interacts with the device again for the target application, the device can directly understand the user's intent based on the loaded summary information and historical interaction information, eliminating the need for the user to repeatedly describe past interactions. This improves interaction continuity and avoids the inefficiency caused by lost historical memory, enhancing the user experience.

[0155] In some exemplary implementations, when the device loads historical interaction information for a target application in response to a user trigger, the processing method may further include: comparing the latest historical summary information with the historical interaction information stored in the third storage layer. If there is no preset correspondence between the latest historical summary information and the historical interaction information, generating historical summary information corresponding to the historical interaction information based on the historical interaction information stored in the third storage layer. Loading the generated historical summary information corresponding to the historical interaction information into the second storage layer.

[0156] Among them, the preset correspondence refers to whether the latest historical summary information can completely and accurately reflect the core content of the historical interaction information, that is, whether the user needs reflected by the latest historical summary information and the historical interaction information are the same.

[0157] Without a pre-defined correspondence, the stored summary information may be outdated or incomplete. Therefore, it is necessary to regenerate or correct the summary information based on historical interaction information to ensure that the summary information corresponds to the historical interaction information. After generating the summary information corresponding to the historical interaction information, it is loaded into the second storage layer for use by the in-vehicle AI agent.

[0158] If there is no pre-defined correspondence, the latest historical summary information read from the third storage layer can be directly loaded into the second storage layer without regenerating it.

[0159] More specifically, comparing the latest historical summary information with the historical interaction information stored in the third storage layer to determine if there is a pre-defined correspondence can include: generating corresponding summary information based on the historical interaction information; determining whether the historical summary information fully covers the core requirements reflected in the summary information, and determining whether there are any omissions, contradictions, or delays in the historical summary information. If the summary information fully covers the core requirements reflected in the summary information, and there are no omissions, contradictions, or delays, then it is considered a pre-defined correspondence.

[0160] If the historical summary information does not fully cover the core needs reflected in the summary information, or if there are omissions, contradictions, or delays in the needs, it indicates that it is not a pre-defined correspondence.

[0161] Therefore, when loading historical interaction information, the consistency between the latest historical summary information and the historical interaction information in the third storage layer can be verified. When the two do not have a preset correspondence, the corresponding historical summary information is regenerated based on the historical interaction information and loaded into the second storage layer. This ensures that the summary information loaded into the second storage layer can completely and accurately reflect the historical interaction content, and avoids the situation where the AI ​​agent in the device misunderstands the user's historical needs due to the mismatch between the summary information and the historical interaction information.

[0162] It is worth noting that in this embodiment, the generation of summary information and preference information can be achieved by training an intelligent agent conversational memory assistant, which generates the information based on historical summary information and interaction information stored in the first storage layer. This intelligent agent conversational memory assistant can generate the latest summary information based on the user's interaction with the AI ​​agent, as well as historical summary information, and extract and maintain the user's preference information. The goal of this intelligent agent conversational memory assistant is to maximize the compression of the context length without losing key information related to "application generation / modification," thereby providing high-value historical information and personalized preferences for subsequent AI agents.

[0163] This intelligent agent conversation memory assistant can be built based on a large language model and trained by developers based on the preference information generation rules and summary information generation rules of the above embodiments. The summary information generation rules can include: the output summary information must include the user's core needs for the target application, key modification history, and key actions completed by the AI ​​agent; invalid information such as small talk and meaningless chatter must be removed; the summary length must be adapted to the vehicle's storage requirements; the number of characters in a single summary must be controlled within a preset threshold; and it must comply with the relevant merging and updating specifications corresponding to the preset incremental merging and updating method in the aforementioned embodiments.

[0164] Specifically, the training data for the intelligent agent conversational memory assistant is a labeled sample dataset from the human-computer interaction field. This dataset contains a large number of interactive dialogue samples created, configured, and modified by users and AI agents for in-vehicle applications, and each sample is labeled with summary information and user preference information. Using a general large language model as the base model, the model is trained through this labeled sample dataset, and the trained model serves as the intelligent agent conversational memory assistant.

[0165] It is worth noting that, regarding the human-computer interaction information processing method of this embodiment, based on the above exemplary implementations, in specific implementation, as a preferred embodiment, it is still […]. Figures 1 to 4 and combined Figure 5 and Figure 6 As shown, it may include, for example, reference: Figure 5 When a user sends a new message on the user interface, the interaction information is added to the first storage layer when a round of interaction information is generated. If the interaction information stored in the first storage layer is full (exceeding the preset number), the oldest message is automatically discarded. If it is not full, it is automatically stored in the first storage layer.

[0166] Then check if any of the following update sub-conditions are met:

[0167] The first update sub-condition: The interaction information contains keywords such as "add", "modify", "delete", and "don't".

[0168] The second update sub-condition is that no update summary information is triggered after a total of 4 rounds of interaction (4 rounds is an example of a preset round threshold).

[0169] The third update sub-condition: When the user clicks buttons such as "Confirm Request" or "Save Version", the summary information update instruction is triggered.

[0170] The fourth update sub-condition is: idle timeout, for example, 10 minutes of no interaction in the session.

[0171] If any one of the update sub-conditions is met, the preset summary update condition is satisfied, and the summary information update process begins. Specifically, historical summary information, historical preference information, and the interaction information currently stored in the first storage layer are passed to the intelligent agent conversation memory assistant. The intelligent agent conversation memory assistant intelligently understands the interaction information and generates new summary information and preference information.

[0172] Furthermore, the intelligent agent conversation memory assistant also adopts a preset incremental merging update method, merging new content based on historical summary information to avoid redundant processing.

[0173] After generating the latest summary and preference information, the summary information is updated in the second storage layer, for example, by updating the coreSummary field in the second storage layer. Similarly, the preference information can be updated in the userPreference field in the second storage layer.

[0174] Furthermore, if the current summary update is triggered by a user-manually triggered summary information update command or by an idle timeout, the solidifiedSummary field of the third storage layer can be updated to the latest summary information based on the coreSummary.

[0175] Subsequently, when a user saves or confirms a request, the saveSummary() function can be called to synchronize and persist the summary information stored in the third storage layer, the user's preference information stored in the second storage layer, and the preset number of latest interaction information stored in the first storage layer to the local database.

[0176] If the current summary update is not triggered by the user's manual summary information update command, nor by the idle timeout, then wait for the user to save or confirm the request. When the user saves or confirms, the core summary stored in the second storage layer will be persistently stored in the local database (that is, persisted to the third storage layer).

[0177] When it is necessary to load historical interaction information for a target application, such as Figure 6 As shown, the corresponding summary information is first queried from the local database of the third storage layer based on the user identifier, application identifier, and agent identifier.

[0178] If a fixed summary exists, then that fixed summary is loaded into the second storage layer.

[0179] Furthermore, based on the user identifier and application identifier, the system queries the user's preference information from the local database in the third storage layer. If the information exists, it is loaded into the second storage layer.

[0180] It also queries the local database for recent session records, that is, queries the stored historical interaction information, and restores the retrieved historical interaction information to the first storage layer.

[0181] In addition, if there is a time difference between the loaded interaction information and the summary information, the intelligent agent's conversation memory assistant can perform incremental updates based on the loaded interaction information and summary information to determine the new summary information and load the new summary information.

[0182] In the above preferred embodiments, the specific implementation methods of each step can still be found in the descriptions of the above exemplary embodiments, and the beneficial effects brought about by the design of each step in this preferred embodiment can also be found in the descriptions of the above exemplary embodiments.

[0183] The human-computer interaction information processing method in this embodiment adopts the above design. It stores a preset number of the latest interaction information in the first storage layer according to the interaction time, and stores summary information in the second storage layer, thereby compressing and refining historical interaction information. Furthermore, when preset persistence and update conditions are met, the summary information in the second storage layer and the interaction information in the first storage layer are persistently stored. In this way, using summary information to replace the interaction information of historical sessions for storage saves vehicle system storage resources and preserves the long-term context of user and device interactions with the target application.

[0184] In other words, this embodiment only needs to store summary information (extraction of historical interaction information) and a preset number of the latest interaction information, without needing to store all historical interaction information, thereby reducing the occupation of vehicle storage resources and reducing the storage pressure on the device.

[0185] Furthermore, by retaining summary information, this embodiment enables the long-term preservation of contextual information, which also allows the in-vehicle AI agent to accurately understand user intent and improve the user's driving experience.

[0186] An embodiment of the second aspect of this application provides a device with human-computer interaction and AI interaction functions. Specifically, the device can be a transportation device, specifically a vehicle, or it can be an electronic device such as a smartphone, tablet, smart wearable device, smart speaker, or smart home device.

[0187] The device includes a memory and a processor. The memory stores computer program code that executes the scheme of this application, and the memory also has a data storage system, which includes a first storage layer, a second storage layer, and a third storage layer.

[0188] As described in the aforementioned method embodiments, the first storage layer is a short-term session storage layer, used to store a preset number of the latest interaction information according to the interaction time. The second storage layer is a dynamic summary storage layer, used to store summary information generated based on recent interaction information. The third storage layer is a persistent storage layer, implemented using a local database, used to store relevant information for a long time, and will not be described in detail here.

[0189] The processor is used to execute the computer program stored in the memory, and when executing the computer program, it can implement the human-computer interaction information processing method described in the above method embodiments.

[0190] The device in this embodiment stores a preset number of the latest interaction information in a first storage layer according to the interaction time, and stores summary information in a second storage layer to compress and refine historical interaction information. Furthermore, when preset persistence and update conditions are met, the summary information in the second storage layer and the interaction information in the first storage layer are persistently stored. In this way, using summary information to replace the interaction information of historical sessions for storage saves the device's information storage resources while preserving the long-term context of user interactions with the device for the target application.

[0191] In other words, this application only needs to store summary information (extraction of historical interaction information) and a preset number of the latest interaction information, without needing to store all historical interaction information, thereby reducing the occupation of the device's information storage resources and reducing the storage pressure on the device.

[0192] The above are merely some embodiments of this application and are not intended to limit this application. The technical features or structures in the foregoing different embodiments can be arbitrarily combined to form other specific technical solutions as needed. For those skilled in the art, this application can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the protection scope of the claims of this application.

Claims

1. A method for processing human-computer interaction information, applied to a device with a data storage system, characterized in that, The processing method includes: When a user interacts with the device for a target application, the generated interaction information is stored in the first storage layer of the data storage system. The first storage layer is used to store a preset number of the latest interaction information according to the interaction time. If the preset summary update conditions are met, the summary information of the interaction information currently stored in the first storage layer is determined, and the summary information is stored in the second storage layer of the data storage system. Under the condition of satisfying the preset solidification update, the summary information stored in the second storage layer and the interaction information currently stored in the first storage layer are stored in the third storage layer of the data storage system.

2. The method for processing human-computer interaction information according to claim 1, characterized in that, The conditions for satisfying the preset summary update conditions include: The interaction information generated by the user's interaction with the device includes preset adjustment keywords; or... The number of times the summary information stored in the second storage layer has not been updated has reached a preset round threshold; or, Receive the summary information update instruction triggered by the user; or, The idle time between the user and the device reaches a preset duration.

3. The method for processing human-computer interaction information according to claim 1, characterized in that, The conditions for satisfying the preset solidification and update conditions include: Received a save operation command triggered by the user for the target application.

4. The method for processing human-computer interaction information according to claim 1, characterized in that, When storing the digest information stored in the second storage layer to the third storage layer, the processing method includes: The summary information stored in the second storage layer is bound to the user's user identifier and the application identifier of the target application; The summary information, which is bound to the user identifier and the application identifier, is stored in the third storage layer.

5. The method for processing human-computer interaction information according to claim 1, characterized in that, The step of determining the summary information of the interaction information currently stored in the first storage layer includes: Extract the historical summary information generated by the user for the target application from the second storage layer; Based on the historical summary information and the interaction information stored in the first storage layer, a preset incremental merging and updating method is used to determine the summary information of the interaction information currently stored in the first storage layer.

6. The method for processing human-computer interaction information according to claim 5, characterized in that, The preset incremental merging and updating methods include: If the interaction information stored in the first storage layer is a preset invalid type of information, the historical summary information is determined as the summary information of the interaction information currently stored in the first storage layer. If the interaction information stored in the first storage layer is a preset conflict type information that conflicts with the historical summary information, the historical summary information is corrected according to the interaction information stored in the first storage layer to obtain the summary information of the interaction information currently stored in the first storage layer. If the interaction information stored in the first storage layer is a preset new requirement type information, the historical summary information is supplemented according to the interaction information stored in the first storage layer to obtain the summary information of the interaction information currently stored in the first storage layer.

7. The method for processing human-computer interaction information according to claim 1, characterized in that, The processing method further includes: If the preset summary update condition is met, the user's current preference information is determined based on the interaction information stored in the first storage layer and the user's historical preference information stored therein. If the preset fixed update conditions are met, the user's current preference information is bound to the user's user identifier and then stored in the third storage layer.

8. The method for processing human-computer interaction information according to claim 1, characterized in that, When the device responds to the user trigger and loads historical interaction information for the target application, the processing method further includes: The latest historical summary information of the user for the target application is located in the third storage layer, and the latest historical summary information is loaded into the second storage layer. The user's historical interaction information for the target application is read from the third storage layer and loaded into the first storage layer.

9. The method for processing human-computer interaction information according to claim 8, characterized in that, The processing method further includes: Compare the latest historical summary information with the historical interaction information stored in the third storage layer; If there is no preset correspondence between the latest historical summary information and the historical interaction information, historical summary information corresponding to the historical interaction information is generated based on the historical interaction information stored in the third storage layer. The generated historical summary information corresponding to the historical interaction information is loaded into the second storage layer.

10. A device, characterized in that: The device includes a memory and a processor; The memory has a data storage system, which includes a first storage layer, a second storage layer and a third storage layer, and the memory stores computer programs. When the processor executes the computer program, it implements the human-computer interaction information processing method according to any one of claims 1-9.

Images