A feedback processing method and system based on source weight and a computer readable storage medium

Abstract

In various intelligent systems, the system needs to adjust its behavior based on external feedback. For example, industrial equipment adjusts its operating parameters based on operator instructions, recommendation systems optimize recommendations based on user feedback, and vehicle systems execute operations based on driver instructions. However, existing technologies suffer from homogenized feedback processing, static weights, simplistic feedback handling, and a lack of closed-loop optimization. Therefore, a method is needed that can dynamically adjust weights based on the source, resist interference through an accumulation mechanism, and continuously optimize the feedback processing strategy to improve the robustness and intelligence level of the system. This application provides a feedback processing method, system, and computer-readable storage medium based on source weights, aiming to achieve robust and adaptive feedback processing by dynamically maintaining source weights, accumulating feedback until a threshold is triggered, and continuously optimizing based on the feedback results.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and information processing technology, and more specifically, to a feedback processing method, system, and computer-readable storage medium based on source weights. Background Technology

[0002] During the operation of intelligent systems, the system needs to continuously receive external feedback to optimize its behavior. Typical application scenarios include industrial equipment adjusting operating parameters based on operator instructions, recommendation systems optimizing content pushes based on user feedback, and vehicle systems responding to driver operation commands. However, current feedback processing mechanisms have significant flaws: all feedback sources are treated indiscriminately, failing to differentiate the reliability differences between different sources. For example, instructions from experienced and novice operators are executed equally, easily leading to safety hazards; feedback weights are generally set with fixed values, lacking the ability to dynamically adjust based on historical interaction data, causing the system to be unable to adapt to changes in source reliability; feedback processing mechanisms rely excessively on immediate responses to single feedback, lacking a cumulative verification mechanism, meaning a single erroneous feedback can trigger incorrect actions; and the system lacks the ability to learn from feedback results, failing to feed back historical processing effects into subsequent decision-making processes to form a closed-loop optimization. These flaws make existing systems vulnerable in complex environments, making it difficult to balance response speed and decision accuracy, especially in scenarios with high reliability requirements such as industrial control and vehicle systems, where erroneous feedback interference is particularly prominent. Existing technologies urgently need improvement to address these issues. Summary of the Invention

[0003] The purpose of this application is to provide a feedback processing method, system, and computer-readable storage medium based on source weights, which can distinguish the reliability of different feedback sources, dynamically adjust the weights, avoid the immediate impact of a single erroneous feedback, and thus improve the decision-making accuracy and robustness of the system.

[0004] This application provides a feedback processing method based on source weights, the technical solution of which is as follows: Includes the following steps: Maintain dynamic relationship weights between each feedback source and the target system; Receive feedback from the target system; The effective weight of the feedback is calculated based on the relation weight, where the effective weight increases as the relation weight increases; Cumulative effective weights; When the accumulated weights meet the predetermined conditions, the behavior of the target system is adjusted.

[0005] Furthermore, this application also proposes that accumulation includes accumulation based on feedback source and feedback type.

[0006] Furthermore, this application also proposes that feedback sources include, but are not limited to, at least one of the following: user, intelligent agent, external system, device, or internal module of the system.

[0007] Furthermore, this application proposes that the dynamic relationship weights be updated based on at least one of the following: interaction frequency, historical accuracy, collaboration records, role permissions, degree of sharing, or corrective behavior.

[0008] Furthermore, this application proposes that the relationship weight update includes positive increment and negative decay, and is limited to a preset range.

[0009] Furthermore, this application also proposes that the feedback includes at least one of positive feedback, negative feedback, and correction.

[0010] Furthermore, this application also proposes that the predetermined conditions include the cumulative negative feedback weight exceeding a threshold.

[0011] Furthermore, this application also proposes that adjusting the behavior of the target system includes, but is not limited to, adjusting the target system's response strategy, updating relational data, or reducing its confidence on relevant topics, at least one of these.

[0012] Furthermore, this application also proposes to include: applying time decay to the accumulated weights so that the influence of recent feedback is greater than that of long-term feedback.

[0013] Furthermore, this application also proposes that the time decay adopts a preset decay model to periodically decay the accumulated weight.

[0014] Furthermore, this application also proposes to include storing feedback events and changes in relation weights in a storage module for subsequent analysis and learning.

[0015] Furthermore, this application also proposes that the target system includes, but is not limited to, intelligent agents, industrial control systems, vehicle systems, Internet of Things platforms, recommendation systems, search engines, or online service platforms.

[0016] Furthermore, this application also proposes that, as one implementation method, the method can be deployed on a local device or a cloud server.

[0017] Furthermore, this application also proposes a feedback processing system based on source weights, comprising: The relationship weight maintenance module is used to maintain the dynamic relationship weight between each feedback source and the target system; The feedback receiving module is used to receive feedback to the target system; The weight calculation module is used to calculate the effective weight of the feedback based on the relationship weight, where the effective weight increases as the relationship weight increases; The accumulation module is used to accumulate effective weights; The trigger module is used to adjust the behavior of the target system when the accumulated weights meet predetermined conditions.

[0018] Furthermore, this application also proposes that the accumulation module be configured to accumulate effective weights separately according to the feedback source and feedback type.

[0019] Furthermore, this application also proposes to include a storage module for storing feedback events and changes in relationship weights.

[0020] Furthermore, this application also proposes to include a decay module for time decay of the accumulated weights.

[0021] Furthermore, this application also proposes that the relationship weight maintenance module be configured to update the relationship weight based on at least one of the following: interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, or corrective behavior.

[0022] Furthermore, this application also proposes that the triggering module be configured to trigger an adjustment when the accumulated negative feedback weight exceeds a threshold.

[0023] Furthermore, this application also proposes adjustments including, but not limited to, adjusting the target system's response strategy, updating relational data, or reducing its confidence level on relevant topics.

[0024] Furthermore, this application also proposes that, as one implementation method, the system can be deployed on a local device or a cloud server.

[0025] Furthermore, this application also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0026] As can be seen from the above, the feedback processing method, system, and computer-readable storage medium based on source weight provided in this application solve the problems of indiscriminate feedback processing, static weights, and single-feedback dependence in the prior art by maintaining dynamic relationship weights, calculating effective weights, accumulating weights, and adjusting system behavior when conditions are met. It can distinguish the reliability of different feedback sources, dynamically adjust weights, and avoid the immediate impact of single erroneous feedback, thereby improving the decision-making accuracy and robustness of the system.

[0027] Comparative analysis with existing technologies: Regarding the weighting mechanism: existing technologies use fixed weights or simple static priorities; while the solution in this application can achieve dynamic weights, which are updated in real time based on factors such as historical interactions and accuracy.

[0028] Regarding feedback processing: existing technologies generally process feedback only once and respond immediately; while the solution in this application adopts an accumulation mechanism, which triggers adjustment only after multiple identical feedbacks accumulate to a threshold.

[0029] Regarding source differentiation: Existing technologies may differentiate sources (such as primary / secondary operators); while the solution in this application can achieve multi-dimensional differentiation, accumulating data separately according to source and type. Regarding learning ability: existing technologies lack learning capabilities; however, the solution proposed in this application can achieve closed-loop learning, and the feedback results are used to update the source weights.

[0030] In terms of technical effectiveness: existing technologies are susceptible to interference from false feedback and cannot adapt to changes; while the solution of this application can achieve anti-interference, self-adaptation, and continuous optimization.

[0031] The above comparison shows that this application solves the inherent defects of existing feedback processing systems through a complete mechanism of "dynamic weighting + cumulative triggering + closed-loop learning". Attached Figure Description

[0032] Several embodiments of this application are described below with reference to the accompanying drawings. It should be noted that the specific structures, modules, steps, parameters, and connections shown in the drawings are preferred embodiments of this application and not limitations on the scope of protection of this application. Those skilled in the art can make various modifications, substitutions, or combinations to the specific details shown in the drawings based on the teachings of this application, and these modified embodiments should still be considered to fall within the scope of protection of this application.

[0033] Figure 1 This application provides a system architecture diagram, which illustrates an exemplary architecture of the feedback processing system of this application. Each module can be adjusted according to actual applications and does not limit the scope of protection.

[0034] Figure 2 This is a flowchart of a feedback processing method provided for this application. The diagram is for illustrative purposes only and does not constitute a limitation on the claims. Detailed Implementation

[0035] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. Other technologies that may be mentioned in the embodiments can be implemented using existing technology or other patent applications filed by the applicant on the same day, and will not be repeated here. It should be particularly noted that the specific module divisions, process steps, data flow directions, status names, time values, etc., shown in the accompanying drawings are merely illustrative examples and should not constitute a limitation on the scope of protection of the claims of this application. The scope of protection of the claims is determined solely by their wording and should be interpreted in accordance with the overall content of the specification.

[0036] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0037] In the feedback processing of intelligent systems, several problems arise: Homogeneous feedback processing manifests as treating all feedback sources as equally important, failing to differentiate between sources based on reliability, historical accuracy, or importance; static weighting occurs when feedback weights are set to fixed values, unable to dynamically adjust based on historical interactions; a single feedback processing mechanism results in immediate responses to individual feedback, lacking an accumulation mechanism; and a lack of closed-loop optimization prevents the system from learning from feedback results to optimize future processing strategies. Specifically, homogeneous feedback processing prevents the system from implementing differentiated responses based on source characteristics; static weighting limits the system's dynamic adaptability to historical interaction data; a single feedback processing mechanism allows a single erroneous feedback to directly trigger system actions; and the lack of closed-loop optimization hinders the system's ability to continuously optimize feedback processing strategies. These issues collectively affect the system's robustness, adaptability, and the accuracy of its behavioral adjustments.

[0038] For example, in an industrial control system, the system receives instructions from a primary operator and an assistant operator as feedback input. Instructions from experienced primary operators and novice assistant operators are treated as equally important, with their weights fixed at a static value where the primary operator's weight is higher than the assistant operator's. This weighting cannot be dynamically updated based on the operator's historical interaction frequency, instruction accuracy, and corrective actions. The system responds immediately to individual instructions, and single erroneous instructions are executed directly, lacking a mechanism for accumulating effective weights. The system cannot learn from instruction execution results to adjust its trust level in operators, resulting in historical erroneous instructions not being included in the weight update process. Furthermore, the feedback processing mechanism in this scenario cannot distinguish between differences in operator reliability; a single erroneous feedback may directly trigger unexpected equipment actions. Over long-term operation, the system cannot adapt to changes in operator skills, ultimately affecting the safety and stability of equipment operation.

[0039] If the above problems are not addressed, the system will be unable to effectively identify sources of high-reliability feedback, resulting in low-quality feedback being executed in the same way; a single false feedback will directly trigger erroneous actions, reducing the continuity of system operation; the system will be unable to dynamically adjust its processing strategy based on historical interactions, and its ability to adjust behavior will gradually degrade over long-term operation; the resulting security risks will accumulate and intensify over time, ultimately damaging the overall reliability of the system.

[0040] To address this, this application proposes a feedback processing method based on source weights, comprising the following steps: Maintain dynamic relationship weights between each feedback source and the target system; Receive feedback from the target system; The effective weight of the feedback is calculated based on the relation weight, where the effective weight increases as the relation weight increases; Cumulative effective weights; When the accumulated weights meet the predetermined conditions, the behavior of the target system is adjusted.

[0041] For ease of understanding, the following explains some key terms in this embodiment: Feedback source: Any entity that provides information or instructions to the target system, such as a human user, other automated system, sensor device, or software module within the system.

[0042] Target system: refers to a system that receives feedback and adjusts its behavior accordingly, such as a smart assistant, industrial control system, vehicle system, recommendation system, etc.

[0043] Dynamic relationship weight: This refers to a quantitative representation of the degree of correlation or trust between the feedback source and the target system. This weight is not fixed but is adjusted in real time based on historical interactions and specific rules.

[0044] Feedback: refers to the information received by the target system from the feedback source. This information can be evaluations, suggestions, instructions, or data regarding the target system's behavior.

[0045] Effective weight: refers to the actual impact of a single feedback instance over a cumulative process. This weight is calculated based on the dynamic relationship weights of the feedback sources; the higher the relationship weight, the greater the effective weight of the feedback.

[0046] Cumulative weights: These refer to the sum of the effective weights from one or more feedback sources over a period of time. By accumulating these weights, the persistence and strength of the feedback can be assessed.

[0047] Predefined conditions: These refer to specific thresholds or states that trigger behavioral adjustments in the target system. When the accumulated weight reaches or exceeds this condition, the target system will be instructed to perform the corresponding behavioral adjustments.

[0048] Adjusting the behavior of the target system: refers to the response measures taken by the target system after it meets predetermined conditions based on the accumulated weights, such as changing its operating parameters, updating internal data, modifying decision logic, or changing the way it interacts with external entities.

[0049] This application provides a feedback processing method based on source weights, the main technical features of which include the following aspects: First, a dynamic relationship weight between each feedback source and the target system is maintained. This application provides a mechanism for establishing and continuously updating the relationship weight between each feedback source and the target system. This relationship weight reflects the target system's level of trust in the feedback source or the importance of its feedback. For example, an initial weight value can be assigned to each feedback source, which can be adjusted in subsequent interactions. One implementation is that the system can maintain a simple counter to record the number of feedbacks from each source, using this as the initial basis for the relationship weight. Another implementation is that the initial priority of different sources can be manually set, for example, marking certain specific sources as "high priority," with their weight values ​​higher than other sources. These weight values ​​can be stored in an accessible database or in-memory structure for use in subsequent steps.

[0050] Secondly, it receives feedback from the target system. During the operation of the target system, it continuously receives information from various feedback sources. This feedback can be user input, sensor data, signals from other systems, or status reports from internal modules. For example, in a smart home system, it can receive feedback from a user via voice command to "turn on the lights," or receive data from a temperature sensor indicating that "the current temperature is 25 degrees Celsius." This feedback information is captured and passed to subsequent processing modules. Feedback can be received in various ways, such as through network interfaces, message queues, API calls, or direct function calls.

[0051] Next, the effective weight of the feedback is calculated based on the relationship weight, where the effective weight increases with the relationship weight. Upon receiving feedback, the system calculates the actual impact of this feedback, i.e., the effective weight, by combining the current relationship weight of the feedback source. The calculation method ensures that the higher the relationship weight of the source, the greater the effective weight generated by its feedback. For example, the received feedback value can be simply multiplied by the relationship weight of the feedback source to obtain the effective weight of this feedback. If the feedback itself is a Boolean value (e.g., "yes" or "no"), it can be converted into a numerical effective weight based on the relationship weight. This calculation method ensures that feedback from different sources is no longer treated equally, but rather receives differentiated influence based on the closeness or trust level between the feedback source and the target system.

[0052] Subsequently, effective weights are accumulated. After calculating the effective weight of a single feedback, the system does not immediately adjust the target system; instead, it accumulates these effective weights. The accumulation process can be a simple summation of all effective weights. For example, the system can maintain one or more accumulation counters, adding each effective weight to the corresponding counter after receiving feedback and calculating it. This accumulation mechanism helps smooth out fluctuations that may arise from a single feedback, preventing unnecessary or erroneous actions by the target system due to accidental or isolated feedback. The scope of accumulation can be global accumulation across all feedback or accumulation targeting specific types or sources of feedback.

[0053] Finally, when the accumulated weight meets a predetermined condition, the target system's behavior is adjusted. The system continuously monitors the state of the accumulated weight. Once the accumulated weight reaches or exceeds a pre-set condition, the target system is triggered to execute corresponding behavioral adjustments. For example, the predetermined condition could be a fixed numerical threshold; when the accumulated weight exceeds this threshold, the feedback signal is considered strong and sustained enough to warrant a response from the target system. Behavioral adjustments can take various forms. For instance, if the accumulated weight indicates a negative trend, the target system might issue a warning; if it indicates a positive trend, the target system might optimize its internal parameters. This cumulative trigger-based adjustment mechanism ensures that changes in the target system's behavior are thoroughly validated and considered, rather than being an immediate reaction to a single feedback.

[0054] The following example will provide a more detailed explanation of the above technical solution: Suppose there exists an intelligent industrial control system for managing robotic arms on a production line. This system receives instruction feedback from different operators and needs to adjust the robotic arm's operating parameters accordingly. Traditional systems may treat all operator instructions the same or process them only according to fixed access levels, which could lead to production accidents caused by errors from inexperienced operators.

[0055] To address this, this application proposes a feedback processing method based on source weights. Specifically: First, the intelligent industrial control system maintains a dynamic relationship weight between each operator (i.e., the feedback source) and the robot arm control module (i.e., the target system). Initially, all operators may be assigned a medium relationship weight. For example, operator A's initial relationship weight is 0.5, and operator B's initial relationship weight is also 0.5.

[0056] Next, the system continuously receives instruction feedback from the operator to the robotic arm. For example, operator A issues a "stop" command, and operator B issues a "accelerate" command. These commands are captured by the feedback receiving module.

[0057] The system then calculates the effective weight of each operator's instruction feedback based on their current relationship weight. Assume operator A has a good operational history, and their relationship weight has increased to 0.8 after multiple correct operations. When operator A issues a "stop" instruction, the effective weight of this instruction is calculated as the base value of the instruction multiplied by 0.8. Operator B, on the other hand, has some erroneous operations in their operational history, and their relationship weight has decreased to 0.3. When operator B issues an "accelerate" instruction, the effective weight of this instruction is calculated as the base value of the instruction multiplied by 0.3. Therefore, instructions from experienced and trusted operators have greater influence.

[0058] The system then accumulates these calculated effective weights. For example, the system maintains an accumulation counter for "stop" commands and an accumulation counter for "accelerate" commands. The effective weight of operator A's "stop" command is added to the "stop" command accumulation counter, and the effective weight of operator B's "accelerate" command is added to the "accelerate" command accumulation counter. If operator A issues multiple "stop" commands consecutively, or if other high-weight operators also issue "stop" commands, the accumulated weight of the "stop" commands will increase rapidly.

[0059] Finally, when the cumulative weight of a certain instruction meets a predetermined condition, such as the cumulative weight of the "stop" instruction exceeding a preset emergency stop threshold, the intelligent industrial control system will adjust the behavior of the robot arm and execute an emergency stop operation. If the cumulative weight of operator B's "accelerate" instruction fails to reach the preset acceleration threshold for an extended period, the robot arm will not immediately execute an acceleration operation, thus avoiding potential risks caused by instructions from a single low-trust source. In this way, the system can achieve robust adjustments to the target system's behavior based on the dynamic trust level of the feedback source, combined with an accumulation mechanism.

[0060] Based on the examples of the intelligent industrial control systems described above, the source weight-based feedback processing method proposed in this application demonstrates significant technological contributions.

[0061] Traditional industrial control systems often suffer from homogenized feedback processing when handling operator commands, meaning all operator commands are treated the same or based solely on static permission levels. For example, regardless of operator A's experience, their commands may have the same initial impact as those of novice operator B. In contrast, this application achieves differentiated feedback processing by maintaining dynamic relationship weights between each feedback source and the target system and calculating the effective weight of feedback based on these relationship weights. In the example above, operator A receives a higher relationship weight due to their good historical performance, and their "stop" command has a significantly higher effective weight than operator B's "accelerate" command. This ensures that feedback from high-trust sources has a greater impact, effectively solving the problem of feedback homogenization.

[0062] Furthermore, weights in existing systems are typically static and cannot be dynamically adjusted based on an operator's historical performance. For example, once an operator is assigned a certain permission level, their instruction weight remains fixed. This application addresses this issue by employing a "dynamic relationship weight" maintenance mechanism, enabling relationship weights to be updated in real-time based on the operator's actual interactions and historical accuracy. In the example, operator A's relationship weight increases due to good performance, while operator B's relationship weight decreases due to misoperation. This allows the system to adaptively adjust its trust level towards different operator instructions, thus resolving the problem of static weights.

[0063] Furthermore, existing systems typically respond immediately to single feedback, lacking a cumulative mechanism, which can lead to erroneous actions from a single false feedback. For example, a novice operator B's accidental "accelerate" command might immediately cause the robot arm to accelerate. This application introduces a cumulative triggering mechanism through the technical means of "accumulating effective weights." In the example, even if operator B's "accelerate" command is assigned an effective weight, it needs to accumulate multiple times to reach a predetermined condition before triggering the robot arm's acceleration behavior. This significantly enhances the system's anti-interference capability, avoids the risks caused by single false feedback, and solves the problem of single feedback processing.

[0064] Finally, traditional systems often fail to learn from feedback results after processing, hindering their ability to optimize future handling strategies for feedback from the same source, thus lacking closed-loop optimization capabilities. This application establishes a closed loop in feedback processing by adjusting the behavior of the target system when the accumulated weights meet predetermined conditions. In the example, when the accumulated weights trigger the robot arm's adjustment behavior, the adjustment result can, in turn, influence the operator's relationship weight update. For instance, if an operator's instruction leads to optimization of system behavior, its relationship weight may be further increased. This mechanism enables the system to continuously learn and optimize its feedback processing strategy, improving the system's intelligence level and solving the problem of lacking closed-loop optimization.

[0065] In summary, this application, through its complete technical concept of dynamically maintaining source weights, calculating effective weights based on weights, and accumulating effective weights until conditions are met to trigger adjustments, effectively solves the problems of homogeneous feedback, static weights, single feedback processing, and lack of closed-loop optimization in existing feedback processing systems, and significantly improves the robustness, adaptability, and intelligence of the system.

[0066] In some of the solutions described above in this application, accumulation is proposed to trigger behavior adjustments when predetermined conditions are met. However, in this process, accumulation is global and does not distinguish between different feedback sources and types, making it impossible to accurately accumulate and control feedback from specific sources or types, thus affecting the accuracy and efficiency of feedback processing. Therefore, this application further proposes that accumulation include accumulation based on both feedback source and feedback type.

[0067] Accumulation by feedback source refers to the system maintaining an independent set of cumulative weights for each individual feedback source. A feedback source can be any entity providing feedback to the target system, such as a user, agent, external system, device, or internal system module. This independent accumulation method ensures that the impact of feedback from different sources is independently evaluated and managed, preventing feedback from high-reliability sources from being diluted by noise from low-reliability sources, or vice versa. Implementation methods may include: creating an independent cumulative counter or cumulative weight variable for each identified feedback source in the data structure; or, by adding feedback source identifiers to the accumulated data, grouping and accumulating based on the source during queries or calculations. Accumulation by feedback type refers to the system maintaining separate cumulative weights for different feedback types from the same feedback source. Feedback type refers to the intent or nature expressed by the feedback, such as positive feedback, negative feedback, or correction. This approach captures the semantic information of the feedback more precisely, enabling the system to respond to feedback of different natures in a targeted manner. Implementation methods may include: further subdividing each feedback source into independent cumulative variables for different feedback types (such as positive feedback accumulation, negative feedback accumulation, and correction accumulation); or, when storing cumulative data, including feedback type identifiers in addition to feedback source identifiers, so as to perform type filtering and accumulation in subsequent processing.

[0068] This application's solution addresses the global and imprecise problems of traditional accumulation methods by further refining the accumulation of effective weights by accumulating them separately according to feedback source and feedback type. Specifically, upon receiving feedback to the target system, the system first calculates the effective weight of the feedback based on its source and dynamic relationship weight with the target system. Subsequently, this effective weight is not simply added to a total accumulated value, but is accumulated into specific accumulated variables based on the specific source of the feedback (e.g., whether it comes from user A or user B) and the specific type of feedback (e.g., whether it is positive or negative feedback). For example, the effective weight of user A's negative feedback is accumulated in "user A's negative feedback accumulation," while the effective weight of user B's positive feedback is accumulated in "user B's positive feedback accumulation." This refined accumulation mechanism allows the system to independently track the contribution of each feedback source to different types of feedback, enabling more targeted and accurate decision-making based on data when subsequently determining whether the accumulated weights meet predetermined conditions. For example, behavior adjustments can be triggered only when the cumulative weight of negative feedback from a specific source exceeds a threshold, or when the cumulative weight of positive feedback of a specific type reaches a certain level. This approach avoids interference between feedback from different sources or of different types, ensuring the accuracy and effectiveness of behavior adjustments and significantly improving the accuracy and efficiency of feedback processing.

[0069] As a specific implementation method, a recommendation system is used as an example. In a recommendation system, there are multiple users (feedback sources) who will give feedback (feedback type) of "like" (positive feedback) or "dislike" (negative feedback) to the recommendation results. The system maintains a dynamic relationship weight between each user and the recommendation system. This weight may be updated based on the user's historical interaction frequency, historical accuracy, etc. When user A gives "dislike" feedback to a recommendation result, the system calculates an effective weight based on user A's current relationship weight. This effective weight is not simply accumulated into the sum of all negative feedback, but is specifically accumulated into the variable "user A's accumulated negative feedback". Similarly, when user B gives "like" feedback to another recommendation result, its effective weight is accumulated into "user B's accumulated positive feedback". The system independently monitors these accumulated weights categorized by source and type. For example, only when "user A's accumulated negative feedback" exceeds a certain preset threshold will the system adjust the recommendation strategy for user A, such as reducing the recommendation priority of certain types of content. Conversely, if "user B's accumulated positive feedback" reaches a certain level, the system may strengthen the recommendation of similar content to user B. This approach allows the recommendation system to make more precise and personalized adjustments based on different users' specific preferences and feedback history, as well as the nature of different feedback, thus avoiding the excessive influence of a single user or a single feedback type on the overall recommendation strategy.

[0070] Through the aforementioned technical solutions, the system achieves refined management of feedback. Because the accumulation process distinguishes between feedback sources and types, the system can more accurately identify the feedback tendencies of specific sources and the influence of specific types of feedback, thereby avoiding the cancellation or confusion between feedback from different sources or types. This allows the system to make decisions based on more targeted accumulated data when determining whether predetermined conditions are met, significantly improving the accuracy and effectiveness of behavior adjustments. For example, it can prevent feedback from a few inaccurate sources from interfering with overall judgment, or avoid information distortion caused by the simple superposition of positive and negative feedback. Ultimately, this refined accumulation mechanism enhances the robustness and intelligence of feedback processing, enabling the target system to respond more accurately and efficiently to changes in the external environment.

[0071] In some of the solutions mentioned above in this application, feedback sources are proposed to distinguish feedback from different sources and dynamically adjust the weights. However, in the implementation process, the types of feedback sources are not specified, which may lead to the system being unable to fully cover all possible source entities, such as users, intelligent agents, external systems, etc., thereby affecting the comprehensiveness and adaptability of feedback processing.

[0072] In this regard, this application further proposes that feedback sources include, but are not limited to, at least one of the following: user, intelligent agent, external system, device, or internal module of system.

[0073] Specifically, the feedback source refers to the entity that generates and sends feedback information. This entity can be a human individual who directly inputs instructions or evaluations through a graphical user interface (GUI) or voice interface, or a user who indirectly expresses preferences through behavioral data (such as clicks, browsing, and dwell time). The feedback source can also be a software or hardware entity with a certain degree of autonomous decision-making and learning capabilities, such as other artificial intelligence models or robots working collaboratively with the target system, or a virtual intelligent agent simulating human behavior or environment. Furthermore, the feedback source can refer to other independently operating external systems that exchange data or call functions with the target system, such as third-party services that transmit data and interact with instructions through API interfaces, or collaborative systems that share information through message queues or databases. The feedback source can also refer to physical hardware with specific functions that can generate or receive data, such as IoT devices like sensors and actuators, or industrial control equipment, in-vehicle hardware, etc. Moreover, the feedback source can also be different functional components or subsystems within the target system, such as monitoring modules, diagnostic modules, or different levels of software services or microservices.

[0074] By explicitly recognizing the diverse types of feedback sources, the proposed solution ensures that the system can differentiate between feedback from human users, other intelligent systems, external interfaces, hardware devices, or internal components when maintaining dynamic relationship weights with the target system. When the system receives feedback from these diverse sources, it can calculate the effective weight of that feedback based on the type of that specific source and its dynamic relationship weight with the target system. This mechanism allows the system to more accurately identify and distinguish the differences in reliability, importance, or historical accuracy of feedback provided by different entities, thereby avoiding processing biases caused by a single feedback source type. For example, feedback from high-privilege users may be assigned a higher initial weight, while data from unauthenticated external systems may require more stringent verification or a lower weight. This nuanced differentiation and processing makes the accumulation of effective weights and subsequent adjustments to the target system's behavior more rational and effective.

[0075] The following example illustrates this concept. In a smart home system, the target system is the overall smart home control center. This system needs to process information from various feedback sources to adjust its behavior. For example, a user might issue a command like "raise the air conditioner temperature" via a voice assistant (the user), which is one feedback source. Simultaneously, an energy-optimizing agent within the smart home system (the agent) might suggest "adjust the indoor temperature to 25 degrees Celsius" based on historical data and current environmental sensor data (devices), which is another feedback source. Furthermore, the system might receive feedback from external weather forecast services (external systems), such as "the temperature is expected to rise in the next hour," and from internal device health monitoring modules (internal modules), such as "the air conditioner compressor is operating at too high a load." For these different feedback sources, the smart home system maintains dynamic relationship weights for each. For example, the user's direct command might have a higher weight, the energy-optimizing agent's suggestion weight might be dynamically adjusted based on its historical energy-saving performance, the external weather service's weight might be based on its forecast accuracy, and the device health monitoring module's weight might be based on the accuracy and importance of its diagnoses. When the system receives these feedbacks, it calculates and accumulates effective weights based on the dynamic relationship weights of their respective sources. When the accumulated weights meet predetermined conditions, such as when the accumulated negative feedback weights exceed a threshold, the system will adjust its behavior, such as reducing the air conditioning load or issuing a warning to the user.

[0076] Through the above technical solution, this application solves the problem of insufficient specification of feedback source types, thereby significantly improving the comprehensiveness and adaptability of feedback processing. The system can cover and process feedback from various entities, including human users, intelligent agents, external systems, devices, and internal system modules, ensuring that the feedback mechanism has stronger robustness and flexibility in complex and ever-changing application scenarios. This refined distinction of feedback sources enables the system to more accurately evaluate and utilize the value of information from different sources when maintaining dynamic relationship weights and accumulated effective weights, avoiding decision-making biases that may result from a single source type, and thus enhancing the intelligence level and reliability of the entire feedback processing system.

[0077] In some of the embodiments described above in this application, dynamic relationship weights are proposed to adjust the processing based on the feedback source. However, in its implementation, it is necessary to specifically define how to update the weights based on the interaction history and attributes to ensure that the weights can truly reflect the reliability and importance of the source and avoid static issues.

[0078] In this regard, this application further proposes that dynamic relationship weights be updated based on at least one of the following: interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, or corrective behavior.

[0079] Dynamic relationship weight refers to a numerical indicator established by the system for each feedback source and the target system, which changes over time, through interaction, and performance. This weight reflects the degree of influence, trustworthiness, or importance of a specific feedback source on the target system. It can be implemented as a floating-point or integer value that fluctuates within a preset minimum and maximum range, such as [0, 1] or [0, 100], where higher values ​​represent higher trust or influence. Another implementation method is to use hierarchical weights, dividing feedback sources into different trust levels, each corresponding to a weight range, and allowing sources to dynamically migrate between different levels. Interaction frequency refers to the number or density of interactions between a specific feedback source and the target system, used to measure the activity and participation of the feedback source. It can be implemented by recording the number of times a feedback source submits feedback or engages in other forms of interaction with the system per unit of time (e.g., daily, weekly) and adjusting the relationship weight based on this number. Historical accuracy refers to the correctness, effectiveness, or consistency with the actual operating results of the system based on the feedback or instructions provided by the feedback source in the past. The implementation can be achieved by comparing the historical feedback from the feedback source with the actual processing results of the system or external verification results to calculate its accuracy, error rate, or deviation value. Collaboration records refer to the historical performance of the feedback source in completing tasks or collaborating with other entities (such as other users, agents, or system modules). This can be achieved by recording the number of collaborative tasks the feedback source participated in, the success rate of collaboration, the role played in the collaboration, and the evaluations of other collaborators. Role permissions refer to the preset permission level or role assigned to the feedback source in the target system or its organization. This can be achieved by presetting different base weights or weight adjustment coefficients for different roles (such as administrator, advanced user, and ordinary user). Sharing degree refers to the depth, breadth, or openness of information sharing between the feedback source and the target system. This can be achieved by evaluating the completeness and timeliness of the information proactively provided by the feedback source or the openness of its data interfaces. Corrective behavior refers to the behavior of the feedback source proactively correcting, withdrawing, or providing remedial measures after discovering its own or the system's errors. This can be achieved by recording the number of error corrections proactively submitted by the feedback source, the timeliness of the corrections, and the effects of the corrections.

[0080] This application's solution addresses the problem of static weights in traditional feedback processing by maintaining a dynamic relationship weight between each feedback source and the target system, and updating this weight based on multiple factors. Specifically, when the system receives feedback from a specific source, it doesn't simply treat all feedback the same; instead, it first evaluates the dynamic relationship weight between the feedback source and the target system. This dynamic relationship weight is not static but is adjusted in real-time based on the historical interaction performance between the feedback source and the target system. For example, sources with high interaction frequency will have their weight maintained or increased due to activity; sources with high historical accuracy will have their weight increased due to reliability; if a source performs well in collaboration, its collaboration records will positively influence its weight; sources with higher role permissions naturally have higher initial weights or greater influence; sources with high sharing levels have more flexible weights due to their openness; and sources that can promptly correct their behavior can effectively avoid significant weight drops due to occasional errors. By integrating these multi-dimensional factors, the system can construct a comprehensive, detailed, and adaptive profile of feedback sources. When any or a combination of these factors changes, the dynamic relationship weight will be adjusted accordingly. This dynamic adjustment mechanism allows the system to flexibly assign different influences to feedback sources based on their actual performance and importance. It no longer simply responds to every feedback instance, but rather ensures that only feedback from trustworthy, active sources with a good historical track record can effectively drive the target system's behavioral adjustments when accumulated to predetermined conditions.

[0081] As a specific implementation method, the processing of operator instructions in industrial equipment is used as an example. In an industrial control system, the system maintains a dynamic relationship weight for each operator, which reflects the operator's level of trust in the system. Specifically, when operator A sends an instruction to the system, the system updates its dynamic relationship weight based on the following factors: The system records the number of times operator A has sent instructions over a period of time. If operator A sends instructions frequently, its relationship weight may increase slightly, reflecting the impact of interaction frequency. The system tracks whether operator A's past instructions have caused equipment malfunctions or required subsequent corrections. If operator A's instructions repeatedly cause the equipment to enter an abnormal state, its historical accuracy score will decrease, thus significantly reducing its relationship weight. Conversely, if operator A's instructions consistently and effectively optimize the production process, its historical accuracy score will increase, and its relationship weight will increase accordingly. If operator A has collaborated with other operators to complete complex operations with a high success rate, the system considers operator A to have good collaborative abilities, thus positively impacting its relationship weight. Assuming operator A is a senior engineer with higher operational authority, the system will set a higher base relationship weight for them to reflect their professionalism and authority. If operator A realizes that their instruction is erroneous and promptly retracts it or issues a corrective instruction, the system will record this corrective action. This may, to some extent, offset the weight decrease caused by the erroneous instruction, or even slightly increase the weight to reward their responsible attitude. Through this mechanism, operator A's dynamic relationship weight will be updated in real time based on their performance in these dimensions. For example, an experienced operator with high historical accuracy and high privileges will maintain a high relationship weight; while a novice operator may have a lower initial weight and need to gradually increase it through multiple accurate operations. When the system receives an instruction from operator A, it calculates the effective weight of the instruction based on their current dynamic relationship weight and accumulates it. Only when the accumulated effective weight meets predetermined conditions will the system execute the corresponding equipment adjustment action.

[0082] Through the above technical solution, this application effectively addresses the problem of static weights in traditional feedback processing. By dynamically updating relationship weights based on multiple factors such as interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, or corrective behavior, the system can more accurately and meticulously assess the reliability and importance of each feedback source. This allows the system to assign differentiated influence to feedback sources based on their actual performance and authority, rather than simply treating them all the same. For example, feedback from high-credibility sources can influence system behavior more quickly, while feedback from low-credibility sources requires stronger accumulation to trigger adjustments. This dynamic and adaptive weight adjustment mechanism significantly enhances the system's resistance to erroneous feedback, improves the accuracy and robustness of feedback processing, and enables the system to continuously learn and optimize trust strategies for different feedback sources, thereby improving the overall level of intelligence.

[0083] In some embodiments described above, dynamic relationship weight updates are proposed to adjust weights based on factors such as interaction frequency and historical accuracy. However, in this process, weights may grow or decay indefinitely, lacking a constraint mechanism, leading to unstable system behavior or weight values ​​exceeding a reasonable range, thus affecting the reliability of feedback processing. To address this, this application further proposes that relationship weight updates include positive increments and negative decays, limited to a preset range.

[0084] Relationship weight update refers to the process of adjusting the relationship weight corresponding to a feedback source based on the interaction between the feedback source and the target system. This process can be based on predefined update rules or algorithms. For example, the magnitude of the weight increase or decrease can be determined by evaluating the nature of the feedback event (such as accuracy and validity); alternatively, it can be achieved by integrating machine learning models that learn from historical feedback data and adaptively adjust update strategies to more accurately reflect the reliability of the feedback source. Positive increment refers to increasing the relationship weight when the feedback source exhibits positive, accurate, or expected behavior during the relationship weight update process. For example, when feedback provided by a feedback source is verified as correct or valid by the system, its relationship weight can be slightly increased; or, when there are frequent and successful collaboration records between the feedback source and the target system, a positive increment can reflect an increase in trust. Negative decay refers to decreasing the relationship weight when the feedback source exhibits negative, erroneous, or unexpected behavior during the relationship weight update process. For example, when feedback from a source is verified as erroneous or invalid by the system, its relationship weight can be reduced to a certain extent. Alternatively, when a feedback source has not interacted with the target system for a long time, or its historical accuracy declines, its relationship weight can be gradually reduced through negative decay to reflect its weakening influence. A preset range refers to setting a clear upper and lower limit for the relationship weight, ensuring that the value of the relationship weight always remains within these limits. For example, the relationship weight can be limited to a numerical range, such as [0, 1] or [0, 100], where 0 represents the lowest trust level and 1 or 100 represents the highest trust level. Alternatively, different trust level ranges can be set according to specific application scenarios, such as "low trust," "medium trust," and "high trust," and a corresponding weight range can be set for each level to prevent the weight from growing or decaying indefinitely, thereby maintaining the stability and predictability of the system.

[0085] This application's solution combines a relationship weight update mechanism with preset range limits to ensure the stability and reliability of dynamic relationship weights during the adjustment process. Specifically, when the system receives feedback from a specific feedback source, it determines whether to positively increment or negatively decrease the relationship weight of that feedback source based on the nature of the feedback (e.g., positive, negative, or corrective) and the historical performance of that feedback source. For example, if the feedback is judged to be positive or accurate, the relationship weight is positively incremented; if the feedback is judged to be negative or incorrect, it is negatively decreased. After each increment or decrease operation, the system immediately checks whether the updated relationship weight exceeds the preset range. If the updated weight exceeds the preset upper limit, it is adjusted to the upper limit value; if it is below the preset lower limit, it is adjusted to the lower limit value. This mechanism ensures that relationship weights do not grow or decrease indefinitely, thereby avoiding system behavior instability caused by extreme weight distribution. In this way, the above method can not only dynamically adjust the relationship weights based on factors such as interaction frequency and historical accuracy, but also effectively prevent the weight values ​​from exceeding a reasonable range, so that the system can maintain the robustness and controllability of its feedback processing in the process of continuous learning and adaptation to feedback sources.

[0086] The following example illustrates this. In a smart assistant system, the system needs to adjust its behavior based on user feedback. The system maintains a dynamic relationship weight for each user, reflecting the system's level of trust in that user's feedback. Initially, the relationship weight for all users can be set to 0.5. The system sets the preset range for relationship weights to [0.1, 1.0]. When a user expresses satisfaction with a response from the smart assistant (e.g., saying "Thank you, assistant" via voice), the system recognizes this as positive feedback and increments the user's relationship weight positively, for example, by 0.05. If a user expresses dissatisfaction with the assistant's response or points out an error (e.g., saying "You're wrong" via voice), the system recognizes this as negative feedback and decreases the user's relationship weight negatively, for example, by 0.1. After each increment or decrease, the system checks the updated weight. If the updated weight exceeds 1.0, it is capped at 1.0; if it is below 0.1, it is capped at 0.1. For example, if a user frequently provides accurate feedback, their weight may gradually increase from 0.5 to 1.0 and remain at that value; if a user frequently provides incorrect feedback, their weight may gradually decrease from 0.5 to 0.1 and remain at that value.

[0087] Through the above technical solution, this application effectively solves the problem of infinite growth or decay of relation weights during dynamic updates, thereby avoiding system instability or unreliable feedback processing caused by extreme weights. This solution ensures that relation weights are always maintained within a reasonable and controllable preset range, enabling the system to continuously and adaptively adjust based on factors such as the interaction frequency and historical accuracy of feedback sources, while also guaranteeing the stability and predictability of the adjustment process. This not only improves the system's ability to handle differentiated feedback sources but also significantly enhances the robustness of the entire feedback processing mechanism, ensuring that the system can make more accurate and reliable decisions when faced with complex and ever-changing feedback inputs.

[0088] In some of the solutions described above in this application, feedback is proposed to receive and process input. However, in this process, the feedback is not distinguished by type, which makes it impossible to differentiate the processing for different feedback intentions, affecting the system's response accuracy and efficiency.

[0089] In this regard, this application further proposes that feedback includes at least one of positive feedback, negative feedback, and correction.

[0090] Specifically, positive feedback represents recognition, satisfaction, or support for the target system's behavior or results. This can be achieved through user clicks of "like," high ratings, explicit positive text comments, or confirmation signals sent by internal system modules. Negative feedback represents dissatisfaction, rejection, or problems with the target system's behavior or results. This can be achieved through user clicks of "dislike," low ratings, explicit negative text comments, reports, or error reports and anomaly alerts sent by internal system modules. Correction provides corrective information, aiming to directly correct erroneous behaviors or data in the target system. This can be achieved through user-submitted error reports with correct data, operator manual modification of system parameters, intelligent agents providing alternative solutions, or automatic correction instructions from internal system modules based on verification rules.

[0091] This solution addresses the problem of homogenized feedback processing by clearly defining specific feedback types, enabling the system to respond differently based on the intent of the feedback. Specifically, feedback includes positive feedback, which indicates approval or affirmation of the target system and can be used to enhance relevant behaviors or weights; negative feedback, which indicates negation or problems and can be used to trigger adjustments or corrections; and corrections, which provide corrective information and can be directly used to improve errors. By including at least one of these types, the system can flexibly adapt to different scenarios, avoiding treating all feedback the same, thereby improving processing accuracy and efficiency.

[0092] Upon receiving feedback to the target system, the system can calculate the effective weight of the feedback based on its specific type and the dynamic relationship weights maintained for each feedback source. For example, when positive feedback is received, its effective weight accumulates positively, helping to enhance the confidence of the target system's current behavior or related parameters. When negative feedback is received, its effective weight accumulates negatively; when the accumulated negative feedback weights meet predetermined conditions, adjustments to the target system's behavior will be triggered. When corrective feedback is received, the system can influence the target system's behavior adjustment more preferentially or more directly, depending on its corrective nature, such as directly updating relationship data or revising the response strategy. This differentiated processing based on feedback type allows the system to more accurately understand the feedback intent and, combined with the reliability of the feedback source, make more refined and effective decisions.

[0093] The following example illustrates this. In a recommendation system, when a user interacts with the recommendations, the system receives different types of feedback. For example, a user clicking "like" or "favorite" a recommended product is considered positive feedback. A user clicking "dislike" or "report" a recommended product is considered negative feedback. When a user discovers an error in the inventory information of a recommended product and proactively submits the correct inventory status, this is considered a correction. The system maintains dynamic relationship weights for each user based on factors such as historical interaction frequency and accuracy. When receiving these different types of feedback, the system calculates the corresponding effective weights based on the user's relationship weights and the feedback type. For example, positive feedback from a high-trust user accumulates with a higher effective weight, thus reinforcing the recommendation model's preference for that type of product more quickly; conversely, negative feedback accumulates with a higher effective weight, triggering adjustments to the recommendation strategy more quickly. For corrective feedback, the system may decide, based on the user's relationship weights, whether to immediately update the product database or perform verification first.

[0094] Through the above technical solution, the system can differentiate its processing based on the specific intent of the feedback, avoiding the accuracy and efficiency problems caused by treating all feedback the same. This ability to distinguish between positive, negative, and corrective feedback allows the system to consider not only the reliability of the source but also the nature of the feedback when processing feedback from different sources, thus making more accurate and practically relevant responses. This significantly improves the system's adaptability and robustness to changes in the external environment, ensuring the effectiveness and timeliness of system behavior adjustments.

[0095] In some of the solutions described above in this application, predetermined conditions are proposed to trigger adjustments to the target system's behavior. However, in this process, the predetermined conditions are not specified, and it cannot be ensured that the adjustment is triggered only when the accumulated negative feedback weight reaches a sufficient level. This may lead to the system overreacting to a single false feedback or adjusting untimely, reducing overall robustness and adaptability.

[0096] In this regard, this application further proposes predetermined conditions, including that the cumulative negative feedback weight exceeds a threshold.

[0097] In this context, predetermined conditions refer to a set of preconditions or rules used during system operation to determine whether specific behavioral adjustments need to be triggered. It provides a clear basis for system decision-making, ensuring that adjustments to system behavior are based on preset logic rather than arbitrary triggering. Predetermined conditions can be a simple Boolean expression, such as "cumulative negative feedback weight > threshold," or a more complex logical combination, such as "cumulative negative feedback weight > threshold AND number of negative feedback sources > N," or a judgment based on a time window, such as "cumulative negative feedback weight > threshold in the past X minutes." Cumulative negative feedback weight refers to the total amount of negative feedback received by the system from one or more feedback sources over a period of time, or for a specific set of events, after weighted processing, for a specific target system or its specific behavior. This cumulative value reflects the intensity and persistence of negative feedback. Cumulative negative feedback weight can be obtained through simple summation, i.e., adding its effective weight directly to the current cumulative value each time negative feedback is received, or it can be calculated using weighted averaging or exponential smoothing to give greater influence to recent feedback, or assign different cumulative coefficients to different types of negative feedback. A threshold is a preset critical value used in the system decision-making process to compare the cumulative negative feedback weights and trigger behavioral adjustments. When the cumulative negative feedback weights reach or exceed this threshold, the system will perform corresponding adjustment operations. The threshold can be a fixed value, preset by the system designer based on experience or experimental results; it can be dynamically adjustable, for example, adjusted in real time based on the current operating state of the target system, environmental context, or historical data analysis results; or it can be hierarchical, i.e., setting multiple thresholds, each corresponding to a different degree of system behavior adjustment.

[0098] The proposed solution ensures higher stability and accuracy in handling negative feedback by preferentially setting the predetermined condition for triggering adjustments to the target system's behavior as an accumulated negative feedback weight exceeding a threshold. Specifically, the system first maintains dynamic relationship weights between each feedback source and the target system and receives feedback to the target system. When negative feedback is received, the system calculates the effective weight of this negative feedback based on the dynamic relationship weight of the feedback source. This effective weight is then accumulated to form a cumulative negative feedback weight. This accumulation process effectively filters out single, accidental, and potentially inaccurate negative feedback, preventing unnecessary or erroneous adjustments by the system due to individual erroneous feedback. Only when this accumulated negative feedback weight reaches or exceeds a pre-set threshold does it indicate that the intensity and persistence of the negative feedback have reached a level requiring system intervention. At this point, the triggering module issues an instruction to adjust the target system's behavior. This mechanism cleverly combines the reliability of the feedback source (reflected by dynamic relationship weights), the nature of the feedback (negative feedback), and the persistence of the feedback (reflected by accumulation and threshold judgment), forming a robust and adaptive feedback processing closed loop. By combining with the dynamic relationship weights in the basic scheme, this scheme can ensure that negative feedback from high-trust sources has a greater impact, thereby reaching the threshold more quickly during the accumulation process. This allows the system to respond more promptly to important negative information from reliable sources, while maintaining a certain "tolerance" for negative feedback from low-trust sources to avoid oversensitivity.

[0099] As a specific implementation, in an industrial control system, the system can maintain a dynamic trust weight for each operator. This weight reflects the operator's historical operational accuracy and experience level. When an operator issues a command that may lead to system risk (e.g., negative feedback), the system calculates the effective negative feedback weight of the command based on the operator's current trust weight. For example, a negative feedback issued by a senior operator with a trust weight of 0.9 might have an effective weight of 0.9 multiplied by the original negative feedback value; while the same negative feedback issued by a novice operator with a trust weight of 0.3 might only have an effective weight of 0.3 multiplied by the original negative feedback value. These effective negative feedback weights are continuously accumulated. The system can set a threshold, for example, to trigger the system to enter a safe mode or issue an alarm when the accumulated negative feedback weight reaches 100. If a senior operator has a high effective negative feedback weight, it may only take a few operations to make the accumulated weight exceed the threshold; while for a novice operator, more negative feedback accumulation is required to reach the same threshold. Furthermore, this threshold can be dynamically adjusted according to the current operating status of the industrial equipment. For example, when the equipment is operating under high load, the threshold can be appropriately lowered to improve the system's responsiveness.

[0100] Through the above technical solution, the system effectively addresses the problem of unclear predetermined conditions, ensuring that adjustments to system behavior are based on sufficient and reliable negative feedback. Specifically, by setting a trigger condition where the cumulative negative feedback weight exceeds a threshold, the system significantly improves its anti-interference capability, avoiding erroneous actions caused by single, accidental, and potentially inaccurate negative feedback. Furthermore, since the negative feedback weight is calculated based on the dynamic relationship weights of the sources, negative feedback from high-trust sources will have a greater impact, enabling the system to respond more promptly and accurately to key negative information from reliable sources, thereby enhancing the system's robustness and adaptability. This mechanism provides the system with a quantifiable and adaptive decision-making criterion, making the system more stable and accurate in handling negative feedback, thus optimizing the target system's behavior adjustment strategy.

[0101] In some of the embodiments described above in this application, the behavior of the target system is adjusted to respond to accumulated feedback in order to optimize the system. However, in the implementation process, the adjustment behavior may lack specific implementation details, resulting in a single adjustment method that is not flexible enough and cannot be optimized in multiple dimensions for different feedback types and scenarios. For example, it is impossible to effectively adjust the response logic, update data, or reduce the dependence on erroneous topics, thereby limiting the adaptability and learning effect of the system.

[0102] In this regard, this application further proposes that the behavior of adjusting the target system includes, but is not limited to, at least one of: adjusting the response strategy of the target system, updating relational data, or reducing its confidence on relevant topics.

[0103] Adjusting the target system's response strategy refers to the rules, logic, or algorithms by which the target system processes input, generates output, or executes tasks. Specifically, the system can dynamically modify the priority of its internal decision tree or rule engine based on the type and intensity of feedback. For example, for high-weight negative feedback, the system may switch to a more conservative response mode. Furthermore, the system can adjust the behavior patterns of its external interfaces. For instance, for user feedback, the system may change the response style of its chatbot, the order of information presentation, or the preference settings of its recommendation algorithm. Updating relational data refers to the data stored internally by the target system that describes the relationships or attributes between entities. Specifically, the system can directly modify the dynamic relational weights related to the feedback source based on the feedback results. For example, if feedback from a source is verified as accurate, its relational weight increases; if it is verified as incorrect, its relational weight decreases. Simultaneously, the system can update data in its internal knowledge graph or database related to specific topics, entities, or behaviors. For example, based on user feedback on a product, the "popularity" attribute of that product can be updated. Reducing its confidence on a related topic refers to the target system's level of trust or certainty regarding a specific topic, information, or judgment. Specifically, the system can adjust the parameters related to the topic in its internal model, making it more cautious about information on that topic in subsequent processing. For example, it can reduce the weight of a certain knowledge point in the question-answering system. In addition, the system can mark the topic as "unverified" or "low credibility" and attach warning messages or seek more evidence when outputting externally or making internal decisions.

[0104] This application's solution addresses the lack of detail in behavioral adjustment by providing multiple specific adjustment methods. When the accumulated weights meet predetermined conditions, the system no longer responds in isolation but selectively executes operations such as adjusting the target system's response strategy, updating relational data, or reducing its confidence level on relevant topics, based on the nature of the feedback and the dynamic relational weights of the source. This mechanism enables the system to achieve deeper optimization from basic feedback reception, weighting, and accumulation. For example, when negative feedback from highly reliable sources accumulates to a certain level, the system not only identifies the need for adjustment but can also precisely avoid future errors on specific topics by reducing confidence or correcting internal knowledge by updating relational data. This refined adjustment capability allows the system to more intelligently adapt to changes in the external environment, achieving more targeted self-optimization and learning, thereby significantly improving the system's robustness and intelligence.

[0105] The following is a concrete example. In an intelligent assistant system, the assistant maintains dynamic relationship weights with each user and receives user feedback on the assistant's actions. When a user provides negative feedback on an assistant's answer (e.g., "What you said is wrong"), the system calculates the effective weight of this negative feedback based on the user's dynamic relationship weights and accumulates it. Assuming the user is a long-term active user with a high historical accuracy rate in feedback, their relationship weight is high, and therefore their effective weight for negative feedback is also high. When the accumulated effective weight of negative feedback reaches a preset threshold, the system triggers behavior adjustment. As a specific implementation, the system can perform the following operations: First, adjust the target system's response strategy. For example, when the intelligent assistant interacts with the user on related topics later, it may switch to a more cautious response mode, or provide additional information about the source or uncertainty when providing information. Second, update relationship data. The system can update entries related to the erroneous information in its internal knowledge base based on this negative feedback, for example, by lowering the credibility rating of the entry or cross-validating it with a more authoritative information source. Third, reduce its confidence level on related topics. The intelligent assistant may adjust the parameters in its internal model related to the erroneous topic, making it less confident in its output when dealing with similar topics in the future, thereby prompting the system to be more rigorous or seek more verification when providing information.

[0106] Through the above technical solution, this application solves the problems of insufficient detail, singular methods, and lack of flexibility in adjusting behavior in traditional feedback processing. The system can selectively execute various refined operations, such as adjusting response strategies, updating relationship data, or reducing confidence on relevant topics, based on the nature of accumulated feedback and the dynamic relationship weights of its sources. This enables the system to perform targeted, multi-dimensional optimizations instead of simply making general adjustments when responding to feedback, thus significantly enhancing the system's adaptability and learning effectiveness. The system can more accurately correct its own behavior, update internal knowledge, and manage the degree of trust in specific information, effectively avoiding insufficient optimization or error propagation caused by singular adjustment methods, thereby improving the overall performance and intelligence level of the system.

[0107] In some of the solutions described above in this application, a cumulative effective weight is proposed to trigger the behavior adjustment of the target system when the cumulative weight meets a predetermined condition. However, in this process, the cumulative weight includes all historical feedback without distinguishing between recent and distant time, which leads to the influence of long-term feedback being equal to that of recent feedback. This may cause the system to respond to current changes with lag or be interfered with by old data, reducing the timeliness and adaptability of feedback processing.

[0108] In response, this application further proposes to decay the accumulated weights over time, so that the impact of recent feedback is greater than that of long-term feedback.

[0109] Time decay refers to the process by which the influence, weight, or value of data or information gradually decreases over time. In feedback processing, this means that earlier feedback information has a smaller impact on the current accumulated weight than newer feedback information. This time decay can be implemented in several ways. For example, an exponential decay model can be used, where all accumulated weights are multiplied by a decay factor less than 1 at each preset time step, causing the contribution of older weights to gradually decrease. Alternatively, a linear decay model can be used, where a fixed value is subtracted from the accumulated weights after each preset time period, or a validity period is set for each feedback event, after which the weight contribution of that feedback event becomes zero or significantly reduced. Furthermore, a half-life model can be used, defining a time period such that the influence of the feedback is halved after the period ends. The core objective of the time decay mechanism is to ensure that the influence of recent feedback is greater than that of distant feedback, aiming to ensure that the system responds more sensitively and accurately to the latest situations and avoids being misled by outdated information. This can be achieved by applying a decay function to the existing accumulated weights each time new feedback is accumulated, so that the weight of the new feedback is directly added to the already decayed old weights, thus naturally reflecting the higher influence of recent feedback. Alternatively, when calculating the total cumulative weight, a timestamp can be attached to each feedback event, and a weight coefficient can be dynamically calculated based on the time difference between the current time and the feedback time. This coefficient decreases as the time difference increases, and then the effective weight of the feedback is multiplied by this coefficient before being accumulated.

[0110] This application's solution optimizes the dynamism and accuracy of feedback processing by introducing a time decay mechanism to process accumulated weights. Specifically, after receiving feedback to the target system and calculating the effective weight of the feedback based on relational weights, the system does not simply accumulate all historical effective weights. Instead, during the accumulation of effective weights, a time decay process is applied to the accumulated weights periodically or with each new feedback. This decay process ensures that the influence of earlier feedback information on the current total accumulated weight gradually decreases over time, while recently received feedback maintains its high influence. It is precisely this dynamic weight adjustment that allows the accumulated weights to more accurately reflect the current feedback situation faced by the target system. Therefore, when the accumulated weights meet predetermined conditions, the triggered adjustments to the target system's behavior can respond more promptly and accurately to the latest system state and external environmental changes. This mechanism effectively solves the problem of equal influence between long-term and recent feedback in traditional accumulation methods, avoiding response delays or decision-making biases caused by interference from old data, thereby improving the timeliness and adaptability of the entire feedback processing system.

[0111] As a specific implementation method, an intelligent customer service system can be considered. This system maintains a dynamic relationship weight for each user and adjusts its service strategy based on the accumulated effective weights from user feedback. To ensure the system can respond promptly to changes in user needs, a time decay mechanism can be employed. For example, the system can set a daily decay factor, such as 0.95. This means that at the end of each day, the accumulated effective weights of all users will be multiplied by 0.95. When a user submits new feedback, the effective weight of that feedback will be directly added to the decayed accumulated weights. For instance, if a user submitted negative feedback three days ago with an effective weight of 10, after two days of decay, their contribution might become 10 * 0.95 * 0.95 ≈ 9.025. A negative feedback submitted today with an effective weight of 10 will be added as a full 10. In this way, recent user feedback has a significantly greater impact on cumulative weight than long-term feedback, enabling the system to prioritize and respond to users' latest emotions and needs. For example, if a user frequently submits negative feedback recently, even if there is a large amount of positive feedback in the past, the system can more quickly identify the user's current dissatisfaction and adjust its service strategy in a timely manner.

[0112] Through the above technical solutions, the system effectively solves the problems of delayed response and interference from outdated data caused by the lack of differentiation between recent and distant data in the cumulative weighting. By applying time decay to the cumulative weights, the system ensures that recent feedback has a higher weighting in the accumulation process, while the influence of long-term feedback gradually weakens over time. This allows the system to prioritize the latest feedback data, avoiding interference from outdated information in the cumulative results, thereby significantly improving the sensitivity of feedback response and the accuracy of decision-making. The system can dynamically adapt to environmental changes, enhancing its overall robustness and self-learning ability, ensuring that behavioral adjustments better meet current practical needs, thus improving the timeliness and adaptability of feedback processing.

[0113] In some of the solutions mentioned above in this application, time decay is proposed to make the impact of recent feedback greater than that of long-term feedback. However, in this process, the specific implementation method of time decay is not defined, which may lead to a lack of systematicness and controllability in the decay process, and make it impossible to ensure the consistency and adaptability of the decay effect, thereby affecting the timeliness and optimization capability of feedback processing.

[0114] In response, this application further proposes to use a preset decay model for time decay, which periodically decays the accumulated weight.

[0115] The pre-defined decay model refers to a predetermined mathematical function or algorithm used to describe the pattern of cumulative weight reduction over time. This model ensures the decay process is systematic, controllable, and consistent, avoiding arbitrariness and making the decay effect predictable. One implementation is a linear decay model, where the cumulative weight decreases by a fixed value after each pre-defined time unit. For example, it decreases by 0.1 weight units every 24 hours. Another implementation is an exponential decay model, where the cumulative weight decays by a fixed proportional factor after each pre-defined time unit. For example, the cumulative weight becomes 90% of its current value every 24 hours. Furthermore, a half-life-based decay model can be used, defining a time period after which the cumulative weight is halved.

[0116] Periodically decaying accumulated weights refers to performing a decay operation on the accumulated weights at predetermined time intervals or triggering conditions. This periodic mechanism ensures the regularity and continuity of the decay, preventing the weights from accumulating indefinitely and causing historical feedback to have an excessive influence, thus ensuring that recent feedback always has priority in influencing system behavior. One implementation method is to set the decay to be based on fixed time intervals; for example, the system can be configured to automatically perform a decay operation at specific times every hour, day, or week. Another implementation method is to set the decay to be triggered by specific events, such as when the system load is low or after daily data synchronization is completed.

[0117] This application's solution addresses the undefined implementation method of time-based decay by introducing a pre-defined decay model and a periodic decay mechanism, ensuring a more systematic, controllable, and efficient decay process. In the aforementioned method, the system accumulates effective weights for feedback. To ensure these accumulated weights accurately reflect the current relevance of the feedback and that the influence of recent feedback is greater than that of long-term feedback, this solution employs a pre-defined decay model. This model defines the specific rules and methods by which accumulated weights decrease over time, avoiding arbitrariness in the decay process and making the decay effect predictable and consistent. Based on this, the system periodically decays these accumulated weights. This periodically executed decay operation ensures the regularity and continuity of the decay, effectively preventing accumulated weights from becoming outdated over time, thus ensuring that the system prioritizes and adopts the latest and most relevant feedback information when adjusting its behavior. In this way, this solution significantly improves the timeliness management of feedback processing and supports the system in continuously optimizing its feedback processing strategy in dynamic environments, thereby enhancing overall robustness and learning efficiency.

[0118] As a specific implementation method, an intelligent assistant system can be considered, which dynamically adjusts its behavior based on user feedback. To ensure that the influence of recent user feedback is greater than that of long-term feedback, the system can employ a preset exponential decay model. For example, a decay factor of 0.9 can be set, meaning that after each decay cycle, the accumulated weight will become 90% of its original value. Simultaneously, the system can be configured to perform a periodic decay operation once daily at midnight. Specifically, after the system receives user feedback, calculates and accumulates the effective weights, at midnight each day, the system automatically iterates through the accumulated feedback weights of all users and updates them according to the preset exponential decay model (i.e., the current accumulated weight multiplied by 0.9). In this way, the accumulated weights of earlier feedback that have not been reinforced by recent feedback will gradually decrease, while recently active feedback will maintain a higher influence, enabling the intelligent assistant to respond to the user's current needs and preferences more promptly and accurately.

[0119] Through the above technical solution, this application effectively solves the problem of the undefined specific implementation method of time decay. The introduction of a preset decay model makes the decay process of accumulated weights have clear rules and predictability, avoiding the arbitrariness of decay operations and thus ensuring the consistency of decay effects. At the same time, the mechanism of periodically decaying accumulated weights ensures the regularity and continuity of decay operations, effectively preventing the infinite accumulation of old feedback weights and their excessive influence, and ensuring that the system can prioritize and adopt recent and more timely feedback. This significantly improves the timeliness and optimization capability of feedback processing, enabling the system to adapt more sensitively to environmental changes and continuously optimize its behavior, thereby enhancing the robustness and adaptability of the system.

[0120] In some of the solutions mentioned above in this application, features such as maintaining dynamic relationship weights and accumulating effective weights are proposed to achieve adaptive feedback processing. However, in this process, the system lacks a persistent storage mechanism for feedback events and changes in relationship weights, which makes it impossible to perform historical data analysis and learning, thereby limiting the closed-loop optimization capability of the system and making it impossible to continuously improve the feedback processing strategy from past interactions, thus affecting the system's adaptability and robustness.

[0121] In response, this application further proposes to store feedback events and changes in relation weights in a storage module for subsequent analysis and learning.

[0122] Specifically, the feedback event refers to a detailed record of each feedback operation received by the system, including key information such as the feedback source, feedback type (e.g., positive feedback, negative feedback, or correction), feedback content, and the time of occurrence. This technical feature records the raw data of the system's interaction with external entities, providing a foundation for subsequent data analysis and learning. The feedback event can be recorded in the form of a structured log, such as storing it in a log file in JSON or XML format; alternatively, the feedback event can be stored in a database, such as creating a dedicated table in a relational or non-relational database to store the attributes of each feedback. The relationship weight change refers to the specific change record of the dynamic relationship weight between the feedback source and the target system during the update process, including the weight value before the update, the weight value after the update, the reason for triggering the update (e.g., an increase in interaction frequency or a decrease in historical accuracy), and the time of the update. This technical feature provides a historical trajectory of weight adjustments, facilitating the tracing and understanding of the logic behind the dynamic weight changes. The relationship weight change can record the old and new values ​​with each update, along with updated metadata in a dedicated weight history table; alternatively, the relationship weight change can be published as a special event through a message queue and subscribed to and stored by a persistent service. The storage module is a component or system for persistently storing data, capable of securely and reliably saving feedback events and relation weight change data. This technical feature ensures data is not lost and is available for long-term access and retrieval. The storage module can employ a local file system, storing data on the server's local hard drive, for example, in CSV or JSONL format; alternatively, it can utilize a distributed database system, leveraging distributed storage services such as Hadoop HDFS or Amazon S3, or distributed databases such as Cassandra or Elasticsearch, to support large-scale data storage and high-concurrency access. The subsequent analysis and learning refers to using the stored feedback events and relation weight change data to perform data mining, pattern recognition, machine learning, and other operations to optimize the system's feedback processing strategy or behavior. This technical feature enables closed-loop learning and continuous improvement of the system, enhancing its adaptability and robustness. The subsequent analysis and learning can include offline batch processing analysis, periodically reading historical data from the storage module, performing statistical analysis using data analysis tools to discover patterns, such as identifying high-reliability or low-reliability feedback sources; or, the subsequent analysis and learning can include machine learning model training, using historical feedback events and corresponding system behavior adjustment results to train a predictive model, such as predicting the future accuracy of a certain feedback source, or optimizing the parameters for weight updates.

[0123] This application's solution stores feedback events and changes in relationship weights in a storage module, enabling the system to comprehensively record its internal state and external interactions throughout the entire process of receiving feedback, calculating effective weights, accumulating weights, and ultimately adjusting the target system's behavior. This persistent storage mechanism gives the previously ephemeral dynamic relationship weights and accumulated effective weights a traceable and analyzable "memory." When the system calculates the effective weight of feedback based on relationship weights and adjusts the target system's behavior based on accumulated weights, every occurrence of a feedback event and the resulting change in relationship weights is precisely recorded. This stored data is no longer just used for immediate decision-making but provides a valuable data foundation for subsequent analysis and learning. Through in-depth mining of this historical data, the system can identify the reliability patterns of different feedback sources, the effectiveness of weight update rules, and the accuracy of behavior adjustments triggered by predetermined conditions. For example, the system can analyze which feedback sources have higher historical accuracy, thereby optimizing their incremental or decay strategies for relationship weight updates; or, the system can learn under what accumulated weight conditions behavior adjustments are most appropriate to avoid misoperation or delayed response. This continuous optimization based on historical data enables the system to learn and improve from past experience, making the maintenance of its dynamic relationship weights more accurate, the strategy for accumulating effective weights more intelligent, and ultimately achieving more robust and adaptive feedback processing.

[0124] The following is a concrete example to illustrate this. In an intelligent recommendation system, the system optimizes its recommendation algorithm based on user feedback (such as liking, disliking, or clicking on recommended content). When a user "dislikes" a recommendation, this constitutes a feedback event. The system records detailed information about this feedback event, including the user ID, the ID of the disliked recommended content, the feedback type (negative feedback), and the time of occurrence. Simultaneously, based on the user's historical feedback performance, the system may adjust the relationship weight of that user as a source of feedback (for example, if the user frequently provides valuable negative feedback, their weight may increase; conversely, if their feedback often fails to meet expectations, their weight may decrease). These changes in relationship weights, including the values ​​before and after the change and the reasons triggering the change, are also recorded. All this data on feedback events and relationship weight changes is stored in a distributed storage module. For subsequent analysis and learning, the system can periodically or in real-time retrieve data from this storage module. For example, the data analysis module can analyze which users' negative feedback led to the most significant improvements in the recommendation model over the past month, thereby further increasing the relationship weights of those users. Alternatively, machine learning models can use this historical data to train a predictive model that can more accurately predict the reliability of future user feedback based on historical feedback patterns. This allows the model to dynamically adjust the parameters for updating relationship weights, enabling the system to more effectively utilize user feedback to optimize recommendation results.

[0125] Through the above technical solution, the system gains persistent storage capability for feedback events and changes in relationship weights, thus solving the problem of lacking historical data analysis and learning mechanisms. This enables the system to continuously learn and improve from past interactions, significantly enhancing its closed-loop optimization capabilities. Therefore, the system can more accurately maintain dynamic relationship weights, more intelligently accumulate effective weights, and ultimately achieve more adaptive and robust feedback processing, effectively improving the system's intelligence level and anti-interference ability.

[0126] In some of the solutions described above in this application, a target system is proposed to receive feedback and adjust behavior. However, in this process, the type of target system is not specifically defined, which may lead to the method lacking specificity in practical applications and failing to adapt to the system requirements of different scenarios. For example, the real-time requirements of industrial control systems for operator feedback differ from the accumulation mechanism of recommendation systems for user feedback, thus affecting the universality and robustness of the feedback processing mechanism.

[0127] In this regard, this application further proposes that the target system includes, but is not limited to, intelligent agents, industrial control systems, vehicle systems, Internet of Things platforms, recommendation systems, search engines, or online service platforms.

[0128] Intelligent agents refer to autonomous entities capable of perceiving their environment, making decisions, and executing actions. They can be rule-based expert systems that reason and make decisions using pre-defined logic and knowledge bases, or systems based on machine learning models that learn patterns through training data and perform predictions or controls. Industrial control systems are automated systems used to monitor and control industrial production processes, such as SCADA systems, DCS systems, and PLC systems. They can consist of programmable logic controllers (PLCs) and human-machine interfaces (HMIs), collecting data through sensors and controlling actuators according to pre-defined programs. They can also be based on a distributed control system (DCS) architecture, where multiple controllers work collaboratively to achieve comprehensive monitoring and control of complex industrial processes. In-vehicle systems are electronic systems installed inside vehicles to provide functions such as driver assistance, infotainment, and vehicle control. They can be IVI (In-Vehicle Infotainment) systems integrating navigation, multimedia playback, and Bluetooth communication, or advanced driver assistance systems (ADAS) that use sensors such as radar and cameras to achieve functions such as adaptive cruise control and lane keeping. An IoT platform refers to a software platform that connects, manages, and processes data from IoT devices. It provides services such as device access, data storage, data analysis, and application development. It can be a cloud-based platform offering device connectivity, message routing, data processing, and storage, or a locally deployed private IoT platform connecting devices via protocols such as MQTT and CoAP, providing data visualization and device management interfaces. A recommendation system is a system that recommends items or information that a user might be interested in based on their historical behavior, preferences, and item characteristics. It can be a system based on collaborative filtering algorithms, analyzing user-item interaction data to discover similarities between users or correlations between items for recommendations, or a system based on deep learning models, using neural networks to learn complex feature representations of users and items to generate personalized recommendations. A search engine is a system that crawls the internet, builds an index, and provides relevant search results based on user queries. It can be a system based on inverted indexes and TF-IDF algorithms, returning search results through keyword matching and relevance ranking, or a system combining machine learning and natural language processing technologies, providing more accurate search results by understanding user query intent and content semantics. An online service platform refers to a platform that provides various services through the Internet, such as e-commerce platforms, social media platforms, and online education platforms. It can be a platform based on a microservice architecture, which decouples different functional modules and communicates through APIs to achieve high availability and scalability. It can also be a platform based on a content management system (CMS), which provides content publishing, user management, and interactive functions.

[0129] The solution proposed in this application maintains dynamic relationship weights between each feedback source and the target system. Upon receiving feedback to the target system, it calculates the effective weight of the feedback based on the relationship weights, and then accumulates these effective weights. When the accumulated weights meet predetermined conditions, the behavior of the target system is adjusted. By specifically selecting intelligent agents, industrial control systems, vehicle systems, IoT platforms, recommendation systems, search engines, or online service platforms, the solution proposed in this application can be optimized and adapted to the characteristics of different types of systems. For example, for industrial control systems, feedback processing may focus more on safety and real-time performance, responding faster to instructions from high-weight operators and requiring more rigorous cumulative verification for instructions from low-weight operators. For recommendation systems, the focus may be more on the accumulation mechanism of user feedback and personalized learning to avoid a single erroneous feedback having too much impact on the recommendation results. This specificity allows the general feedback processing logic to be seamlessly integrated into various practical application scenarios, and its parameters and strategies can be adjusted according to the unique needs of each system, thereby improving the practicality, robustness, and adaptability of the entire feedback processing method.

[0130] The following example illustrates this. Taking a recommendation system as an example, the target system is the recommendation algorithm or model. Feedback sources can be user actions such as clicking, liking, saving, sharing, or expressing disinterest in recommended content. The system maintains a dynamic relationship weight between each user (feedback source) and the recommendation system (target system). This weight can be updated based on the user's historical interaction frequency, click-through rate of recommendations, and the accuracy of feedback. For example, a user who frequently clicks on and purchases recommended products will have a higher feedback weight; while a user who frequently clicks but never purchases may have a lower feedback weight. When a user provides feedback on a recommendation (e.g., liking a product), the recommendation system calculates the effective weight of this feedback based on the user's current relationship weight. High-weighted user feedback will generate a larger effective weight. These effective weights are accumulated, and can be accumulated separately by user and feedback type (e.g., like, dislike). When the accumulated effective weights (e.g., the accumulated weights of positive feedback for a product or a category of products) reach a preset condition (e.g., exceeding a certain threshold), the recommendation system will adjust its behavior, such as updating the recommendation model parameters, increasing the priority of that product or category of products in future recommendations, or adjusting the user's interest profile.

[0131] Through the aforementioned technical solution, this application clarifies the specific type of the target system, enabling the general feedback processing method to be optimized and adapted to the system requirements of different application scenarios. This solves the problem of the method lacking specificity in practical applications, ensuring that the feedback processing mechanism can be effectively applied to various practical systems such as intelligent agents, industrial control systems, vehicle systems, IoT platforms, recommendation systems, search engines, or online service platforms. For example, in industrial control systems, the weight of operator instructions can be dynamically adjusted based on the operator's experience and permissions to ensure the safety of critical operations; in recommendation systems, the influence of feedback can be adjusted based on the user's historical behavior and feedback quality, thereby achieving more accurate personalized recommendations. This greatly enhances the versatility, practicality, and robustness of the feedback processing mechanism, enabling it to better adapt to complex and ever-changing application environments.

[0132] Traditional feedback processing methods often employ relatively fixed deployment methods, making it difficult for the system to flexibly adapt to the privacy and performance requirements of different application scenarios. For example, in privacy-sensitive environments, the system may need to be deployed locally to prevent data leakage; while in high-computational-demand scenarios, cloud deployment is required to fully utilize cloud resources. This limitation restricts the versatility and applicability of feedback processing methods.

[0133] In response, this application proposes a feedback processing method, which, as one implementation method, can be deployed on a local device or a cloud server.

[0134] Specifically, "local devices" refer to computing hardware running near the data source or user, characterized by low processing latency, low data transfer costs, and high data privacy. Local devices can include, but are not limited to, embedded systems, edge computing devices, industrial controllers, smart terminals (such as smartphones and tablets), or private servers within an enterprise. This can be achieved by directly installing the software modules of the feedback processing method on these devices, or by running them in a local environment using containerization technology. "Cloud servers," on the other hand, refer to remote server clusters providing computing services via the internet, characterized by elastic scaling of computing resources, large storage capacity, and the ability to handle large-scale concurrent processing. Cloud servers can include, but are not limited to, virtual machines, container services, or serverless computing services provided by public cloud platforms (such as Amazon Web Services, Microsoft Azure, and Google Cloud Platform), or private cloud environments built by enterprises themselves. This is typically achieved by deploying the various modules of the feedback processing method as microservices on a cloud platform, utilizing cloud services for data storage, computation, and management.

[0135] This application's solution addresses the limitations of feedback processing methods in adaptability across diverse scenarios by providing deployment flexibility, ensuring the system can optimize resource utilization and data security according to actual needs. Specifically, the solution allows for dynamic selection of deployment locations based on application scenario characteristics, such as privacy protection or computing power requirements. When the feedback processing method is deployed on a local device, it can fully utilize local computing resources to achieve low-latency real-time feedback processing, which is particularly suitable for scenarios with strict requirements for response speed and data privacy, such as industrial control systems or vehicle systems. In this deployment mode, the entire process—maintaining dynamic relationship weights for each feedback source, receiving feedback, calculating effective weights, accumulating effective weights, and adjusting the target system behavior when the accumulated weights meet predetermined conditions—is completed on the local device, thereby ensuring that data does not leave the local area and improving system security and response efficiency.

[0136] On the other hand, when the feedback processing method is deployed on a cloud server, it can leverage the powerful computing and storage capabilities of the cloud platform to process large-scale feedback data, supporting complex weight update models and broader adjustments to the behavior of target systems. This is particularly suitable for scenarios requiring the processing of massive amounts of user feedback, complex data analysis, and model training, such as recommendation systems or search engines. In this deployment mode, feedback data can be collected from users or systems globally and sent to the cloud, where cloud servers centrally maintain relationship weights, calculate effective weights, accumulate data, and adjust behavior. Cloud deployment provides elastic scalability to handle peak and trough variations in feedback traffic and supports more complex machine learning models to optimize dynamic updates of relationship weights and behavior adjustment strategies. Therefore, whether deployed locally or in the cloud, this solution effectively supports the operation of the basic feedback processing method and leverages its advantages according to specific needs, making the entire feedback processing mechanism more robust and universal.

[0137] The following is a concrete example. In a smart home system, the system needs to adjust device behavior based on voice commands from family members. Considering the privacy needs of family members and the requirement for real-time response to commands, this feedback processing method can be deployed on local devices such as smart gateways or smart speakers within the home. For example, the smart gateway, as a local device, is responsible for maintaining a dynamic relationship weight between each family member (feedback source) and the smart home system (target system). When a family member issues a voice command (feedback), the smart gateway calculates the effective weight of the command based on the member's relationship weight and accumulates it. When the accumulated weight meets a predetermined condition (e.g., a member repeatedly issues the command "turn off all lights" with a high weight), the smart gateway directly adjusts the behavior of the smart home system, such as turning off all lights. The entire process is completed locally, ensuring the privacy of voice command data and providing an immediate response.

[0138] As another specific implementation, in a large online education platform, the platform needs to optimize course recommendations and platform functions based on learning feedback from global users. Due to the massive number of users, the huge amount of feedback data, and the need for complex machine learning analysis, this feedback processing method can be deployed on cloud servers. For example, the cloud server cluster of the online education platform maintains a dynamic relationship weight between each user (feedback source) and the platform (target system). When users rate, like, or suggest courses (feedback), the cloud server receives this feedback and calculates the effective weight of the feedback based on the user's historical activity, learning outcomes, and other relationship weights. These effective weights are accumulated in the cloud and subdivided by dimensions such as course type and user group. When the accumulated weight meets predetermined conditions (e.g., the negative feedback weight of a course exceeds a threshold), the cloud server adjusts the platform's behavior, such as removing the course, optimizing the recommendation algorithm, or triggering course content review. Cloud deployment enables the platform to elastically handle massive amounts of feedback and utilize powerful computing resources for in-depth analysis and optimization, thereby improving the overall service quality.

[0139] Through the above technical solution, the feedback processing method of this application can be flexibly deployed on local devices or cloud servers according to the needs of actual application scenarios. This deployment flexibility effectively solves the limitations caused by the fixed deployment method of traditional methods, enabling the method to adapt to various environments with different requirements for privacy protection, real-time performance, computing power, and scalability. This not only ensures the efficient operation of the feedback processing method in different scenarios, but also significantly improves the versatility and universality of the entire feedback processing system, enabling it to better serve diverse intelligent systems.

[0140] Traditional feedback processing systems suffer from several problems when handling external feedback, including homogenized feedback processing, static weights, simplistic feedback handling, and a lack of closed-loop optimization. For example, industrial control systems treat all operator commands the same, failing to differentiate between experienced and novice operators; recommendation systems respond immediately to single user feedback, potentially leading to malicious feedback distorting the recommendation model; and vehicle systems lack cumulative verification mechanisms for driver commands, making them susceptible to accidental misoperations. These issues result in insufficient system robustness and difficulty adapting to dynamically changing interactive environments.

[0141] To address this issue, this application proposes a feedback processing system based on source weights. The system maintains a dynamic relationship weight between each feedback source and the target system through a relationship weight maintenance module. This weight is updated in real-time based on factors such as historical interaction frequency, historical accuracy, and collaboration records. For example, when a feedback source provides accurate feedback, its relationship weight increases according to a preset rule; when it provides incorrect feedback or does not interact for a long period, the relationship weight decays according to a preset rule and is limited to the range [0,1]. A feedback receiving module receives feedback to the target system, including at least one of positive feedback, negative feedback, and correction. It captures input information from human users, sensor devices, or other systems through API interfaces or message queues. A weight calculation module calculates the effective weight of the current feedback based on the current relationship weight. The effective weight is positively correlated with the relationship weight, specifically implemented through a multiplicative model: Effective Weight = Feedback Base Value × Relationship Weight, ensuring that feedback from high-trust sources has a greater impact. An accumulation module accumulates the effective weights, maintaining independent accumulation counters for each feedback source and feedback type. For example, different counters can be set for "stop" and "accelerate" commands, and a time decay factor is introduced to make the impact of recent feedback greater than that of long-term feedback. The trigger module continuously monitors the cumulative weight status. When the cumulative negative feedback weight exceeds a preset threshold, it triggers behavior adjustments in the target system, including updating the system response strategy, reducing the confidence of related topics, or performing physical operations.

[0142] The core innovation of this embodiment lies in combining dynamic relationship weight maintenance with an accumulation triggering mechanism in a closed-loop learning manner, thereby achieving differentiated processing and adaptive adjustment of feedback sources. Specifically, the relationship weight maintenance module dynamically updates weights based on historical interaction data, solving the problem of static weights; the weight calculation module ensures that effective weights increase with relationship weights, achieving differentiated feedback processing and avoiding feedback homogenization; the accumulation module requires multiple feedbacks to accumulate to a threshold before triggering adjustments, effectively filtering out single erroneous feedbacks and solving the problem of handling single issues; after the triggering module performs behavior adjustments, it feeds the adjustment results back to the relationship weight maintenance module for subsequent weight updates, forming a closed-loop learning mechanism that enables the system to continuously optimize feedback processing strategies.

[0143] The following explanation uses an industrial equipment control system as an example. The system maintains dynamic relationship weights for operators A and B, both initially at 0.5. Operator A's relationship weight increases to 0.8 due to high historical accuracy; operator B's relationship weight decreases to 0.3 due to multiple erroneous operations. When operator A issues a "stop" command, the weight calculation module calculates the effective weight as the base value of the command × 0.8; when operator B issues an "accelerate" command, the effective weight is the base value of the command × 0.3. The accumulation module adds the effective weight of the "stop" command to a dedicated counter and the effective weight of the "accelerate" command to another counter. The system only executes a stop operation when the accumulated weight of the "stop" command exceeds the emergency stop threshold; operator B's "accelerate" commands require multiple accumulations to reach the trigger threshold. If the system executes a stop and confirms it as a correct operation, the relationship weight maintenance module further increases operator A's relationship weight; if executing acceleration causes an anomaly, it further decreases operator B's relationship weight. Through this mechanism, the system effectively prevents production accidents caused by novice operators' erroneous operations while strengthening its responsiveness to commands from highly reliable operators.

[0144] Through the above technical solutions, the system achieves anti-interference, adaptability, and continuous optimization capabilities. Single false feedback is effectively filtered out by the accumulation mechanism, requiring multiple identical feedbacks to trigger adjustments; dynamic updates of relationship weights enable the system to automatically adjust trust levels based on historical interactions; feedback from high-trust sources has a greater impact, ensuring that critical instructions are executed first; and the closed-loop learning mechanism allows the system to continuously optimize its weight strategy from feedback results. Therefore, this application significantly improves the robustness, adaptability, and intelligence of the feedback processing system, and can be widely applied in various scenarios such as industrial control, vehicle systems, and recommendation systems.

[0145] In some of the embodiments described above in this application, an accumulation module is proposed to accumulate effective weights. However, in its implementation, if the feedback source and feedback type are not distinguished, the accumulated weights may be inaccurate, and it may be impossible to finely process the reliability differences of different sources and the characteristics of feedback types, thereby affecting the robustness, adaptability and optimization effect of the system.

[0146] To address this, this application further proposes an accumulation module configuration that accumulates the effective weights separately based on the feedback source and feedback type. This configuration aims to ensure that the system can identify and distinguish feedback from different entities, thereby reflecting the unique attributes of each source during the accumulation process, such as reliability, permissions, or historical performance. Specifically, an independent accumulation counter or data structure can be allocated in the storage system for each independent feedback source (e.g., through a unique identifier, such as user ID, device serial number, agent name, etc.). Another implementation is to use a mapping table or hash structure, where the key is the identifier of the feedback source and the value is the accumulated weight data corresponding to that source, thereby enabling independent tracking and accumulation of feedback from different sources. Simultaneously, this configuration aims to enable the system to differentiate and process feedback of different natures, such as positive feedback, negative feedback, or corrective feedback, to more finely reflect their impact on the target system. Specifically, within the accumulation data structure of each feedback source, sub-accumulation quantities for different feedback types can be further subdivided (e.g., for a given source, its positive feedback accumulation weight, negative feedback accumulation weight, and corrective feedback accumulation weight can be maintained separately). In addition, multiple independent sets of accumulated weights can be maintained, each set dedicated to a specific feedback type, and the accumulation can be further differentiated within each set based on the feedback source.

[0147] Upon receiving feedback to the target system, the system first determines the effective impact of the feedback based on the dynamic relationship weights maintained by the relationship weight maintenance module and the effective weights calculated by the weight calculation module. Subsequently, the accumulation module no longer accumulates all effective weights uniformly. Instead, it precisely accumulates the effective weights into specific accumulation channels based on the specific source of the feedback (e.g., user A, agent B, external system C) and the feedback type (e.g., positive feedback, negative feedback, correction). For example, the effective weight of user A's negative feedback is accumulated in the "user A - negative feedback" accumulation, while the effective weight of agent B's positive feedback is accumulated in the "agent B - positive feedback" accumulation. When the weight of any specific accumulation channel meets a predetermined condition, the triggering module adjusts the target system's behavior accordingly. This refined accumulation mechanism allows the system to avoid mutual interference between different sources and types of feedback, ensuring the accuracy and relevance of the accumulated weights. For example, a small amount of negative feedback from a high-reliability source may quickly trigger an adjustment, while negative feedback from a low-reliability source requires more accumulation to reach the trigger condition, thus significantly improving the system's accuracy in identifying feedback and the rationality of its response.

[0148] The following is a concrete example to illustrate this. In an intelligent customer service system, there are multiple feedback sources, including users (customers), customer service supervisors (internal experts), and automated evaluation systems (external systems). Feedback types include "satisfied" (positive feedback), "dissatisfied" (negative feedback), and "suggestions for improvement" (correction). When a user gives "dissatisfied" feedback on a service, the feedback receiving module receives this feedback, and the weight calculation module calculates an effective weight based on the user's historical interaction frequency and satisfaction level (relationship weight). The accumulation module then adds this effective weight to the "dissatisfied" feedback accumulation specifically for that user ID. If a customer service supervisor gives "dissatisfied" feedback on a service, because the supervisor's relationship weight is usually higher, their feedback has a larger effective weight, and the accumulation module adds it to the "customer service supervisor - dissatisfaction" accumulation. Since the accumulation for customer service supervisors may have a lower trigger threshold, even a small amount of negative feedback from supervisors can quickly trigger system adjustments, such as training relevant customer service personnel or optimizing service processes. However, user "dissatisfied" feedback may need to accumulate to a higher threshold to trigger similar adjustments. Meanwhile, feedback of the "suggestions for improvement" type, regardless of its source, will be accumulated into the corresponding "suggestions for improvement" accumulation. When a certain threshold is reached, it may trigger the system to optimize functions or update the knowledge base. In this way, the system can perform differentiated and refined processing based on the importance of the feedback source and the nature of the feedback type, ensuring the accuracy and effectiveness of feedback processing.

[0149] Through the above technical solution, the system can independently and accurately accumulate effective weights from different feedback sources and types. This significantly improves the accuracy of weight accumulation, enabling the system to fully consider the reliability differences of different sources and the characteristics of different feedback types, avoiding confusion and dilution of feedback information. Therefore, when determining whether predetermined conditions are met, the system can make more accurate and targeted decisions, thereby effectively improving the robustness, adaptability, and overall optimization effect of the target system, and ensuring the rationality and effectiveness of system behavior adjustments.

[0150] In some of the solutions mentioned above in this application, a system including a relation weight maintenance module, a feedback receiving module, a weight calculation module, an accumulation module, and a triggering module is proposed to handle feedback. However, in this process, there is a lack of a storage mechanism for feedback events and relation weight changes, which makes it impossible to perform subsequent analysis and learning, thus limiting the system's continuous optimization capabilities.

[0151] In this regard, this application further proposes that the system also includes a storage module for storing feedback events and changes in relationship weights.

[0152] The storage module is a component for persistently storing data, recording key data during system operation for subsequent querying, analysis, and learning. This storage module can be implemented in various ways. For example, it can be a database system based on physical storage media such as hard drives or solid-state drives, including relational databases (e.g., MySQL, PostgreSQL) or non-relational databases (e.g., MongoDB, Cassandra). Alternatively, it can be a distributed file system (e.g., HDFS) or a cloud storage service (e.g., AWS S3, Alibaba Cloud OSS) to adapt to different scales and performance requirements. The storage of feedback events and relation weight changes aims to ensure the system can trace historical operations and state evolution, providing a data foundation for system self-learning and optimization. Specifically, feedback events can be stored as structured logs, containing detailed information such as feedback source, feedback type, feedback content, timestamp, and associated target system behavior. Simultaneously, relation weight changes can be recorded as weight update logs, containing key data such as the weight value before the update, the weight value after the update, the reason for the update, and the timestamp. Alternatively, feedback events and relation weight changes can be stored as data records in a time-series database for efficient querying and analysis along the time dimension.

[0153] This application's solution introduces a storage module that works in conjunction with the underlying feedback processing system to form a closed-loop system with memory and learning capabilities. Specifically, when the feedback receiving module receives feedback to the target system, the detailed information of the feedback event is recorded in the storage module in real time. Simultaneously, when the relationship weight maintenance module updates the dynamic relationship weights related to the target system according to preset rules or learning algorithms, these weight change records are also persistently saved by the storage module. In this way, the storage module provides the entire system with a comprehensive historical data view, enabling the system to not only respond instantly but also review past interactions and weight adjustment processes. This stored feedback event and relationship weight change data can be utilized by subsequent analysis modules (e.g., an independent learning module or the optimization logic within the relationship weight maintenance module) to identify feedback patterns, evaluate the effectiveness of weight update strategies, and accordingly optimize the relationship weight update algorithm or adjust the predetermined conditions of the triggering module. For example, the system can analyze historical data to discover that a certain feedback source has high feedback accuracy in a specific context, thereby assigning it a higher initial weight or a faster weight growth rate in the future; conversely, if it is found that the feedback from a certain source frequently causes system malfunctions, its weight decay mechanism can be adjusted. This data-driven optimization process enables the entire feedback processing system to continuously learn and improve from historical experience, thereby enhancing its adaptability and robustness.

[0154] The following is a concrete example. In a smart assistant system, the system includes a relationship weight maintenance module, a feedback receiving module, a weight calculation module, an accumulation module, and a triggering module. For continuous optimization, the system also includes a storage module. This storage module can be an embedded database (e.g., SQLite) or a lightweight file system used to store data locally. When the smart assistant receives user feedback (e.g., a user expressing "satisfaction" or "dissatisfaction" with a reply), detailed information about the feedback event, including the user ID, feedback content, timestamp, and associated assistant replies, is recorded in the storage module. Simultaneously, when the relationship weight maintenance module updates the dynamic relationship weight between the user and the assistant based on factors such as the user's historical interaction frequency and accuracy, the relationship weight changes, including the weight value before the update, the weight value after the update, the reason for the update, and the timestamp, are also recorded. This stored data can be periodically uploaded to the cloud for big data analysis or used for simple statistical analysis locally to optimize the assistant's response strategy or adjust the relationship weight update algorithm, thereby enabling the smart assistant to better understand user intent and provide more accurate services.

[0155] Through the above technical solution, this application effectively solves the problem that existing feedback processing systems lack a storage mechanism for feedback events and changes in relation weights, which prevents subsequent analysis and learning and limits the system's continuous optimization capabilities. Specifically, the introduction of a storage module enables the system to comprehensively record detailed information about each feedback event and the dynamic adjustment process of relation weights, thus providing the system with valuable historical data assets. Based on this historical data, the system can conduct in-depth analysis and learning, identifying the reliability patterns of different feedback sources, the effectiveness of weight update strategies, and the actual effects of system behavior adjustments. This not only supports continuous optimization of the update algorithm for the relation weight maintenance module but also allows the predetermined conditions of the triggering module to be dynamically adjusted according to actual operating conditions, thereby achieving closed-loop optimization of the feedback processing system. Ultimately, the solution of this application significantly improves the system's adaptability and intelligence level, enabling it to continuously learn and improve from historical experience, thus exhibiting higher robustness and superior performance when facing complex and ever-changing external environments.

[0156] In some of the solutions described above in this application, a mechanism for accumulating effective weights is proposed to accumulate feedback weights. However, the weights accumulated in this process may contain outdated feedback information, causing the system to be insensitive to recent changes and unable to prioritize new feedback, thereby affecting the system's adaptability and robustness.

[0157] In response, this application further proposes a feedback processing system based on source weights, which also includes a decay module for time decay of the accumulated weights.

[0158] The decay module is the component in the system responsible for performing time decay operations. This module can be implemented as a software module, for example, by writing specific algorithm code in the system software to process accumulated weights periodically or continuously. Furthermore, in scenarios with high performance requirements, the decay module can also be implemented using dedicated hardware acceleration units, such as using Field-Programmable Gate Arrays (FPGAs) or Application-Specific Integrated Circuits (ASICs) to efficiently perform decay calculations. In a distributed system architecture, the decay module can also be designed as an independent microservice, specifically responsible for receiving accumulated weight data, applying decay logic, and returning the updated weights. Time decay of accumulated weights means that the module can gradually reduce the accumulated weight values ​​over time according to preset rules or models. This time decay can be implemented in various ways. For example, an exponential decay model can be used, where the weight values ​​decrease exponentially with time according to a fixed decay rate; a linear decay model can be used, where the weight values ​​decrease linearly with time according to a fixed step size; or a step-like decay model can be used, where the accumulated weights are significantly decayed at specific time points or periodically. These attenuation mechanisms are designed to ensure that the impact of near-term feedback is greater than that of long-term feedback, thereby enabling the system to respond more sensitively to changes in the current state.

[0159] This application's solution addresses the issue of outdated information in accumulated weights by introducing a decay module that works in conjunction with the basic feedback processing system. Specifically, when the feedback receiving module receives feedback to the target system, the weight calculation module calculates the effective weight of the feedback based on the dynamic relationship weights provided by the relationship weight maintenance module. Subsequently, the accumulation module accumulates these effective weights. Based on this, the decay module periodically or continuously decays the accumulated weights over time. This means that over time, earlier accumulated feedback weights gradually decrease in influence, while more recently accumulated feedback weights maintain a relatively high influence. When the accumulated weights meet predetermined conditions, the triggering module determines whether to adjust the target system's behavior based on the currently decayed accumulated weights. This mechanism ensures that the system prioritizes the latest and most relevant feedback information when making decisions, avoiding decision-making biases caused by the lag in historical feedback. In this way, the system can reflect changes in the current environment and user needs more promptly and accurately, thereby significantly improving the system's adaptability and robustness.

[0160] The following example illustrates this. In a recommendation system, an accumulation module accumulates preference weights for specific content based on a user's historical behavior and feedback. To ensure that recommendations reflect the user's latest changes in interest in a timely manner, the system can be configured with a decay module. This decay module can be a software service that internally implements an exponential decay function. For example, every 24 hours, the decay module applies a decay factor (e.g., 0.9) to all accumulated preference weights, causing older preference weights to gradually decrease. When a user shows interest in new content and provides positive feedback, the effective weight of that feedback is accumulated, and because it is recent feedback, its decay is smaller, thus quickly increasing the priority of that content in the recommendation list. Conversely, if a user showed interest in content a long time ago but has not had any recent interaction, its corresponding accumulated weight will gradually decrease due to time decay, preventing outdated interest from continuing to affect current recommendation results.

[0161] Through the above technical solution, this application effectively solves the problem that accumulated weights may contain outdated feedback information, leading to the system's insensitivity to recent changes. The introduction of the decay module allows the accumulated weights to decay over time, ensuring that the influence of recent feedback is always greater than that of long-term feedback. This enables the system to more sensitively capture the latest trends and changes when evaluating accumulated weights to trigger behavioral adjustments, thereby avoiding decision-making delays or errors caused by the lag of historical data. Therefore, the feedback processing system of this application exhibits higher adaptability, capable of timely and accurate adjustment of the target system's behavior based on dynamically changing external environments and feedback information, significantly improving the overall robustness and intelligence level of the system.

[0162] Traditional feedback processing methods, when maintaining dynamic relationship weights, may lead to inaccurate weight updates or failure to adapt to actual interaction situations if there is a lack of clear update factors. They may also fail to fully reflect the reliability, historical performance, or importance differences of feedback sources, thereby affecting the robustness and adaptability of the system to feedback processing.

[0163] In this regard, this application proposes that the relationship weight maintenance module be configured to update the relationship weight based on at least one of the following: interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, or correction behavior.

[0164] The relationship weight maintenance module is a core component of the system, responsible for managing and adjusting the dynamic relationship weights between various feedback sources and the target system. Its function is to increase, decrease, or correct these weights based on preset rules and real-time data to ensure that the weights accurately reflect the current credibility, importance, or influence of the feedback source. This module can be a standalone software service, such as a microservice, specifically handling weight calculation and update logic; or it can be a functional unit integrated into the main system, performing weight management tasks by calling internal application programming interfaces (APIs). Interaction frequency refers to the number or density of interactions between the feedback source and the target system. A high interaction frequency usually indicates that the source has a deeper understanding of the system or participates more frequently, and its feedback may be more valuable. For example, the system can record the number of times a feedback source submits feedback within a specific time window (such as the past 24 hours or the past 7 days), or record the total amount of data exchange, command sending, and other operations it performs with the system. Historical accuracy refers to the correctness or validity of the feedback provided by the feedback source in the past. A source with high historical accuracy should also have higher credibility in its current feedback. For example, the system can track whether past feedback submitted by a feedback source was ultimately adopted by the system and produced positive results, or whether the reported issues were confirmed as real and resolved. Collaboration history refers to the historical performance of the feedback source in collaborating with the target system or other entities. A good collaboration history may indicate that the source has strong teamwork skills and an understanding of the system's goals. For example, the system can record the number of times the feedback source participated in joint tasks, the role played in collaborative projects, and the degree of contribution to the project's success. Role permissions refer to the preset permission level or identity that the feedback source possesses in the system. Typically, roles with higher permissions (such as administrators or senior engineers) may assign higher initial weight or influence to their feedback. For example, the system can preset different weight bases or weight adjustment caps for different user roles (such as ordinary users, expert users, and system administrators). Sharing level refers to the breadth and depth of information or resources shared between the feedback source and the system. Sources with a high degree of sharing may have a more comprehensive understanding of the system, and their feedback may be more insightful. For example, the system can measure the amount of data provided by the feedback source to the system, the frequency of information updates, or the degree to which they participate in the system's data sharing protocols. Corrective action refers to the proactive measures taken or effective suggestions for improvement by the feedback source after discovering a system problem or error. This behavior demonstrates the source's problem-solving ability and sense of responsibility, and its weight should be positively adjusted. For example, the system can record whether the error report submitted by the feedback source was accompanied by an effective solution, or whether the correction suggestion was adopted by the system and successfully fixed the problem.

[0165] To address the issues of inaccurate weight updates or incompatibility with actual interactions in traditional feedback processing, this application proposes a more refined and dynamic weight maintenance mechanism. The relationship weight maintenance module no longer relies on a single or static factor to evaluate feedback sources, but instead comprehensively considers multiple dimensions, including interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, and corrective actions. When the system receives feedback from a particular feedback source, the relationship weight maintenance module queries and evaluates that source's performance across these dimensions in real time. For example, if a feedback source interacts frequently with the target system, its interaction frequency index will be high; if its historical feedback has been repeatedly verified as accurate, its historical accuracy index will be improved; if it performs well in collaborative tasks, its collaboration records will positively impact its weight; simultaneously, its preset role permissions also provide a base weight. Furthermore, if the source actively participates in information sharing or proactively corrects system errors, these behaviors are also taken into consideration. The relationship weight maintenance module comprehensively analyzes and calculates this multi-dimensional information, for example, through weighted averaging, machine learning models, or preset rule engines, thereby dynamically adjusting the relationship weight between the feedback source and the target system. This multi-factor comprehensive evaluation ensures that the update of relationship weights fully and accurately reflects the current credibility, importance, and contribution of feedback sources to the system. This enables the system to more intelligently identify and respond to high-quality feedback, effectively filtering out low-quality or potentially erroneous feedback, thereby significantly improving the robustness and adaptability of the entire feedback processing system. In this way, the relationship weight maintenance module can provide a more accurate foundation for subsequent effective weight calculation and accumulation, making the entire feedback processing flow more reliable and efficient.

[0166] As a specific implementation method, taking an intelligent customer service system as an example, this system needs to optimize its service strategy based on user feedback. The relationship weight maintenance module maintains a dynamic relationship weight for each user. When a user interacts with the intelligent customer service system, this module comprehensively evaluates the following factors to update the user's relationship weight: First, regarding interaction frequency, the system records the number of conversations between the user and customer service, the frequency of questions asked, etc. For example, an active user who interacts with customer service multiple times a day will have a significantly higher interaction frequency than a user who only interacts once or twice a month. Second, regarding historical accuracy, the system tracks whether the problems reported by the user in the past actually existed and were resolved, or whether their satisfaction rating with the customer service response matches the actual service quality. For example, if multiple system faults reported by a user are confirmed and fixed, their historical accuracy will increase; conversely, if a user frequently submits invalid or misleading feedback, their historical accuracy will decrease. Third, regarding collaboration records, if a user has participated in the system's internal testing, provided valuable suggestions that were adopted, these collaborative behaviors will be recorded and positively affect their weight. In addition, role permissions also affect the initial weight; for example, the initial weight of a VIP customer or system tester may be higher than that of an ordinary user. Regarding the degree of sharing, if a user proactively shares detailed log information, screenshots, or operational steps to help customer service diagnose problems, this indicates a high degree of sharing, which will positively impact their relationship weight. Finally, regarding corrective behavior, if a user provides accurate corrective information or suggestions when they discover incorrect customer service responses, helping the system improve its knowledge base, this behavior will also increase their relationship weight. The relationship weight maintenance module dynamically adjusts the user's relationship weight based on the overall performance of these factors using a preset algorithm (e.g., assigning a weight coefficient to each factor and then performing a weighted sum, or using rule-based fuzzy logic). For example, a VIP user with high frequency, high accuracy, and proactive corrective behavior will have a significantly higher relationship weight than a regular user with low frequency and low accuracy.

[0167] Through the above technical solution, this application effectively solves the problem of inaccurate or unsuitable weight updates in traditional feedback processing. The relationship weight maintenance module dynamically updates relationship weights by comprehensively considering multiple factors such as interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, and corrective behavior. This allows the weights to more comprehensively and accurately reflect the authenticity, credibility, and influence of the feedback source. This avoids the one-sidedness caused by evaluating feedback sources based on a single or static factor, significantly improving the system's ability to differentiate between different feedback sources. For example, for sources with high historical accuracy and frequent interactions, their feedback will receive a higher effective weight, thus influencing the target system's behavior more quickly; while for sources with low historical accuracy or inactivity, their feedback influence will be correspondingly reduced, effectively avoiding interference caused by erroneous feedback. This dynamic, multi-dimensional weight update mechanism gives the entire feedback processing system stronger adaptability and robustness, enabling it to continuously optimize its response strategy based on the actual performance of the feedback source, thereby improving the target system's intelligence level and operational stability.

[0168] In some of the solutions described above in this application, a trigger module is proposed to adjust the behavior of the target system when the accumulated weight meets a predetermined condition. However, in this process, the predetermined condition is not specifically defined, which may lead to the system being unable to effectively distinguish and process different types of feedback, especially the accumulation problem of negative feedback, thereby affecting the robustness and response accuracy of the system. In response, this application further proposes to configure the trigger module to trigger adjustment when the accumulated negative feedback weight exceeds a threshold.

[0169] The trigger module is a component in the system responsible for monitoring the accumulated weights and initiating system behavior adjustments when specific conditions are met. This module can be a software module, such as an independent process, service, or thread, that periodically checks the negative feedback weights in the accumulation module. Alternatively, the trigger module can be an event-driven mechanism, where it is called back to check when the accumulation module updates the negative feedback weights. The accumulated negative feedback weight specifically refers to the value formed in the accumulation module after weighting the negative feedback generated by the target system (such as complaints, error reports, and unsatisfactory evaluations) according to the source weights. This weight can be a numerical variable stored in memory or a database, increasing accordingly each time negative feedback is received and an effective weight is calculated. Furthermore, the accumulated negative feedback weight can also be a time-series data structure, recording the accumulated negative feedback values ​​at different time points for time decay processing. The threshold is a preset critical value used to determine whether the accumulated negative feedback weight has reached the trigger condition. This threshold can be a fixed value, preset by the system administrator or designer based on experience or experimental results. Alternatively, the threshold can be a dynamically adjusted value, adaptively adjusting based on the current state, operating environment, or historical performance of the target system. Triggered adjustment refers to the process of intervening in or changing the behavior of a target system when the accumulated negative feedback weight exceeds a threshold. This can be achieved by calling the target system's application programming interface (API) and sending specific instructions or parameters to change its operating strategy. Alternatively, it can be achieved by sending a signal to the target system's control module to put it into a preset adjustment mode, such as reducing response speed, switching to a backup plan, or issuing an alarm.

[0170] This application's solution addresses the problem of unclear predefined conditions by configuring the trigger module to specifically monitor accumulated negative feedback weights and compare them with preset thresholds. In the entire feedback processing system, the relationship weight maintenance module maintains dynamic relationship weights for each feedback source, the feedback receiving module receives feedback to the target system, the weight calculation module calculates the effective weights of the feedback based on the relationship weights, and the accumulation module is responsible for accumulating these effective weights. Based on this, the trigger module no longer simply checks any type of accumulated weight, but focuses on the accumulation of negative feedback. The trigger module only initiates adjustments to the target system's behavior when the effective weights for negative feedback in the accumulation module reach or exceed the preset threshold. This mechanism ensures that the system does not react immediately to single, accidental, or potentially erroneous negative feedback, but waits for the negative feedback to accumulate to a level sufficient to indicate a real and persistent problem. This allows the system to more accurately identify and respond to genuine negative situations, avoiding unnecessary or erroneous adjustments, thereby improving the overall robustness of the system and the accuracy of decision-making.

[0171] The following is a concrete example. In an industrial control system, a trigger module continuously monitors the cumulative negative feedback weights associated with operator commands. When an operator issues a command (such as an emergency stop command or a command deviating from optimal parameters), the relationship weight maintenance module and the weight calculation module determine the effective weight of the command based on the operator's dynamic trust score (relationship weight). If the command is classified as negative feedback, its effective weight is added to the cumulative negative feedback weights by the accumulation module. The trigger module is specifically configured to check whether this cumulative negative feedback weight, for example, for negative feedback of a specific type of critical command or specific operating parameter, exceeds a preset threshold. For example, if an operator with a low trust score repeatedly issues commands classified as negative feedback, the cumulative negative feedback weight will gradually increase. Only when this cumulative weight exceeds a certain threshold (e.g., indicating a persistent potentially unsafe or incorrect command pattern), will the trigger module initiate adjustments. Such adjustments might include requiring a second operator to confirm critical commands, temporarily locking certain functions for that operator, or issuing an alert to supervisor. The threshold can be a fixed value, such as "50 weighted negative feedback units", or it can be dynamically adjusted according to the current operating mode of the device (e.g., setting a lower threshold during critical operations).

[0172] Through the above technical solution, the system can effectively distinguish and process different types of feedback, especially addressing the issue of accumulated negative feedback. This mechanism avoids interference from single false feedback on system behavior, significantly improving the system's anti-interference capability. Simultaneously, by triggering a condition where the accumulated negative feedback weight exceeds a threshold, the system can more accurately determine when behavioral adjustments are needed, thereby optimizing the timing of adjustments, enhancing the system's robustness and adaptability, and ensuring that the system can make more stable and reliable responses when facing negative inputs.

[0173] In some of the solutions mentioned above in this application, the behavior of the target system is adjusted to respond to the cumulative weight meeting the predetermined conditions. However, in this process, the adjustment behavior may lack specific implementation methods and cannot be precisely optimized for different feedback types or system states, resulting in insufficient flexibility and comprehensiveness in the adjustment of system behavior, which affects the pertinence and effectiveness of feedback processing.

[0174] In response, this application further proposes adjustments including, but not limited to, adjusting the target system's response strategy, updating relational data, or reducing its confidence level on relevant topics.

[0175] Specifically, adjusting the response strategy of a target system refers to dynamically changing the rules, processes, or algorithms of the target system's external interactions or internal processing based on accumulated feedback results. As one implementation method, the system can automatically switch to a more conservative or safer operating mode based on accumulated negative feedback weights; for example, switching from autonomous driving mode to assisted driving mode, or from high-precision recommendation mode to general recommendation mode. As another implementation method, the system can optimize its interaction logic based on accumulated positive feedback weights; for example, an intelligent customer service system can adjust its dialogue flow based on accumulated user satisfaction with specific responses, prioritizing the use of more popular response templates.

[0176] Updating relationship data refers to modifying, adding, or deleting relationship data stored internally or externally based on the results of feedback processing to reflect the latest relationship status. For example, when the historical accuracy of a feedback source accumulates to a certain level, the system can update the historical accuracy record of that source in the relationship weight maintenance module, thereby affecting its dynamic relationship weight. Furthermore, in a multi-agent system, when the collaboration records between two agents accumulate to a certain level, the system can update their collaboration relationship data, for example, by increasing the collaboration frequency count or raising the collaboration trust level.

[0177] Lowering a system's confidence level on a specific topic refers to the system proactively reducing its confidence in providing information, performing operations, or making decisions in a particular domain or topic when it receives accumulated negative feedback. For example, for a knowledge-based question-answering system, if it receives multiple corrections or negative feedback from users on a specific medical topic, the system can lower its confidence level on that topic. In subsequent answers to that topic, it can add disclaimers, prioritize citing authoritative sources, or even refuse to answer. As another example, for a recommendation system, if it repeatedly recommends products that users are not interested in or have given negative reviews of in a specific product category, the system can lower its recommendation confidence level for that category, reducing the recommendation frequency or weight of products in that category.

[0178] This application's solution maintains dynamic relationship weights for each feedback source through a relationship weight maintenance module, and a feedback receiving module receives the feedback. A weight calculation module calculates the effective weight of the feedback based on these dynamic relationship weights, and an accumulation module accumulates the effective weights. When the accumulated weights meet predetermined conditions, a trigger module initiates the behavior of adjusting the target system. Based on this, this application further refines this adjustment behavior, making it no longer a single, general response, but capable of differentiated and refined optimization based on specific feedback types and system states. Specifically, by adjusting the target system's response strategy, the system can dynamically change its external interaction mode or internal processing logic according to the nature and intensity of the accumulated feedback. For example, when the accumulated negative feedback weight indicates that the system is performing poorly in a certain aspect, the system can switch to a more cautious response mode; while when the accumulated positive feedback weight indicates that the system is performing well, its efficiency can be optimized or its functionality expanded. Simultaneously, by updating relationship data, the system can feed the results of feedback processing back to the relationship weight maintenance module, correcting the dynamic relationship weights of the feedback sources in real time. For example, if feedback from a source leads to successful system adjustment, its relationship data can be improved; conversely, if feedback leads to negative results, the relationship data can be reduced. This closed-loop mechanism enables the system to continuously learn and adapt, improving the accuracy of trust assessments of different feedback sources. Furthermore, by reducing confidence in relevant topics, the system can address uncertainties or errors exhibited in specific domains or tasks in a targeted manner. When accumulated feedback indicates a deficiency in the system on a particular topic, the system no longer blindly provides information or performs actions with unwavering confidence, but instead adopts a more conservative strategy, such as seeking further verification, issuing warnings, or temporarily avoiding the topic, thereby effectively preventing potential risks arising from erroneous information or actions. This multi-dimensional adjustment mechanism allows the system to execute more targeted, flexible, and safer behavioral adjustments when it receives accumulated feedback and meets predetermined conditions, overcoming the problems of singular and imprecise adjustment behaviors in traditional feedback processing, and significantly improving the system's robustness, adaptability, and intelligence.

[0179] The following example illustrates this. Consider an intelligent customer service system as the target system, whose task is to handle user inquiries about the "return process." This system maintains dynamic relationship weights for each user, based on factors such as historical user satisfaction and interaction frequency. Suppose that when handling "return process" questions, the system receives multiple negative feedback from users regarding unclear responses and complex processes, and the cumulative effective weight of these negative feedbacks exceeds a preset threshold. At this point, the system will trigger behavioral adjustments. First, the system can automatically adjust its response strategy on the "return process" topic. For example, it might no longer directly provide standardized return process text, but instead prioritize guiding users to a human customer service channel, or add more interactive questions in its responses to clarify the user's specific situation, avoiding directly providing complex information that could cause misunderstanding. Second, the system will update the knowledge base data related to the "return process" topic based on these negative feedback events. For example, it might mark the "unclear" parts mentioned in user feedback and trigger knowledge base maintenance personnel to optimize the content. Furthermore, if a particular response template is frequently marked as "unsatisfactory," the system will lower the internal usage priority of that template and mark it as needing optimization. Finally, given the negative feedback accumulated on the topic of "return process," the system will lower its "confidence" on this topic. This means that the system will be more cautious when handling similar inquiries in the future. For example, it may add disclaimers such as "This information is for reference only; please refer to official regulations for details" when giving a reply, or proactively ask users "Did my answer solve your problem?" after replying to obtain immediate feedback, rather than directly assuming that the problem has been solved.

[0180] Through the above technical solutions, this application provides diverse and refined adjustment options, effectively solving the problems of single and imprecise adjustment behaviors in traditional feedback processing. Specifically, adjusting the target system's response strategy enables the system to dynamically change its behavior patterns based on the nature and intensity of accumulated feedback, ensuring that adjustments can be flexibly implemented for different scenarios and improving the system's adaptability. Updating relational data ensures that the system can correct the dynamic relational weights of feedback sources in real time based on feedback results, forming an effective learning loop and significantly improving the system's adaptability and learning efficiency. Reducing its confidence in related topics enables the system to adopt more cautious strategies when facing uncertain or high-risk areas, effectively avoiding potential risks caused by erroneous information or operations, thereby greatly improving the system's robustness and security. These specific adjustment methods, combined with the basic feedback processing system, enable the system to perform more targeted, flexible, and safer behavioral adjustments when receiving accumulated feedback and meeting predetermined conditions, thereby achieving comprehensive optimization of the target system's behavior and significantly improving the targeting and effectiveness of feedback processing.

[0181] In some of the solutions mentioned above in this application, a system is proposed to implement feedback processing based on source weight. However, in this process, the system deployment lacks flexibility and cannot adapt to the privacy and performance requirements of different scenarios. For example, local processing is required in privacy-sensitive scenarios to protect data, or cloud resources are required in data-intensive scenarios to improve processing efficiency.

[0182] In this regard, this application further proposes that, as one implementation method, the system can be deployed on a local device or a cloud server.

[0183] Specifically, "the system can be deployed on local devices" means that the entire source-weighted feedback processing system, including core components such as the relationship weight maintenance module, feedback receiving module, weight calculation module, accumulation module, and triggering module, can be directly installed and run on the physical hardware owned by the user or organization. This can be achieved by packaging all software components of the system into a standalone application or service and running it directly on a local server, high-performance workstation, or embedded device, thus ensuring that all data storage and processing are completed locally. Alternatively, for resource-constrained edge devices, the system can be deployed as a lightweight module that only processes feedback directly relevant to that device and uploads some aggregated data to the central system when necessary. On the other hand, "the system can be deployed on cloud servers" means that the source-weighted feedback processing system can be deployed on cloud computing infrastructure provided by a third party. This can be achieved by deploying the complete system application on virtual machine instances of cloud service providers such as Alibaba Cloud, Tencent Cloud, Huawei Cloud, AWS, and Azure. Alternatively, cloud platform container services (such as Kubernetes and Docker Swarm) can be used for deployment to achieve high availability and elastic scaling, or serverless architectures (such as AWS Lambda and Azure Functions) can be adopted to deploy each functional module of the system as an independent function and execute it on demand.

[0184] The solution presented in this application provides flexible deployment options, enabling the aforementioned source-weight-based feedback processing system to adapt to diverse application scenarios. When deployed on a local device, the system ensures that sensitive data is processed and stored in the local environment, effectively meeting the needs of application scenarios with strict requirements for data privacy and security, such as industrial control systems or personal health management systems. In this deployment mode, the relationship weight maintenance module can update the relationship weights of local feedback sources in real time, the feedback receiving module can quickly receive local feedback, and the weight calculation module and accumulation module complete the calculation and accumulation of effective weights locally. Finally, the triggering module adjusts the behavior of the target system locally. The entire process does not rely on an external network, ensuring low latency and high security. Conversely, when the system is deployed on a cloud server, it can fully utilize the elastic scalability and powerful computing resources of cloud computing to cope with large-scale, high-concurrency feedback processing needs, such as large-scale recommendation systems or IoT platforms. In this deployment mode, the system can efficiently process data from massive feedback sources. The relationship weight maintenance module can perform more complex weight updates based on large-scale historical data, the accumulation module can handle massive accumulation of effective weights, and the triggering module can adjust the behavior of the target system in a timely manner according to the accumulated weights, thereby ensuring that the system can still operate stably and efficiently under high loads. This deployment flexibility allows the core feedback processing mechanism—dynamic weight maintenance, effective weight calculation, cumulative triggering, and behavior adjustment—to achieve maximum efficiency in the computing environment best suited to its specific needs.

[0185] The following is a concrete example. In a smart home system, a feedback processing system based on source weights can be deployed on a local smart home gateway device. This gateway device, acting as a local unit, is responsible for receiving feedback from family members regarding voice commands, gestures, and other actions taken by smart devices. For example, when a family member issues a "brighten" command to the smart lighting system, the feedback receiving module receives the command, and the relationship weight maintenance module maintains the dynamic relationship weight based on the family member's historical operating habits and permissions. The weight calculation module calculates the effective weight of the command based on this weight, and the accumulation module accumulates it. When the accumulated weight meets predetermined conditions, the module is triggered to adjust the behavior of the smart lighting system. All of this processing is completed on the local gateway, ensuring that family members' privacy data is not uploaded to the cloud, while also ensuring fast command response times. In contrast, for a global online social media platform, its content recommendation system can be deployed on a cloud server. This system needs to handle massive amounts of feedback from hundreds of millions of users worldwide, including likes, shares, comments, and reports. The cloud server provides powerful computing and storage capabilities, enabling the relationship weight maintenance module to maintain dynamic relationship weights for each user, and the feedback receiving module to receive and process high-concurrency feedback streams. The weight calculation and accumulation modules efficiently calculate and accumulate effective weights, while the triggering module adjusts the content recommendation algorithm in a timely manner based on the accumulated weights to optimize user experience. Cloud deployment ensures high availability, scalability, and efficiency in processing large-scale data.

[0186] Through the above technical solutions, this application significantly improves the deployment flexibility of the source-weighted feedback processing system. For privacy-sensitive application scenarios, local device deployment ensures that sensitive feedback data and relationship weights are processed in a controlled local environment, thereby effectively reducing the risk of data leakage and enhancing system security and user trust. For application scenarios that require processing large-scale data or high-concurrency requests, cloud server deployment provides powerful computing resources and elastic scaling capabilities, ensuring that the system can operate efficiently and stably, responding to and processing massive amounts of feedback in a timely manner, thereby improving the system's processing efficiency and availability. This ability to select the deployment environment according to specific needs enables the feedback processing mechanism of this application to be applied more widely and effectively to various complex real-world scenarios, thereby maximizing its technical advantages in anti-interference, adaptability, and continuous optimization.

[0187] In some of the solutions described above in this application, a feedback processing method based on source weights was proposed to address the problems of feedback homogenization, static weights, single processing method, and lack of learning ability. However, when these methods are deployed in computer systems, there is a lack of effective executable forms, leading to implementation complexity and difficulty in updating, which affects the widespread application and efficiency of the methods. To address this, this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the feedback processing method based on source weights.

[0188] Specifically, computer-readable storage media refers to a physical carrier capable of storing computer programs or data, from which a processor can read and execute instructions. This media can be various forms of non-transitory storage devices. For example, it can be a hard disk drive (HDD) for long-term storage of large amounts of program code and data; a solid-state drive (SSD) offering faster read / write speeds and better shock resistance; read-only memory (ROM) or flash memory for storing firmware or embedded system programs; or an optical disc (such as a CD-ROM, DVD, or Blu-ray disc) or a Universal Serial Bus (USB) flash drive for easy program distribution and transfer. These media can persistently preserve computer programs, ensuring they remain accessible even after power loss.

[0189] A computer program is a collection of instructions that, when executed by a processor, enables a computer to perform a specific task or operation. This program can consist of source code written in a high-level programming language (such as Java, Python, C++, etc.), which, after compilation or interpretation, forms machine code or bytecode that can be directly executed by the processor. This program encapsulates all the logic of a source-weight-based feedback processing method, including data structure definitions, algorithm implementations, and interface calls, thereby transforming abstract method steps into a concrete, executable sequence of instructions.

[0190] A processor is the core component of a computer system, responsible for executing instructions in a computer program, performing data processing, and controlling operations. This processor can be a central processing unit (CPU), such as the Intel Core series or AMD Ryzen series, possessing powerful general-purpose computing capabilities; it can also be a graphics processing unit (GPU), used to accelerate specific tasks in scenarios requiring parallel computing; or it can be a microcontroller (MCU), digital signal processor (DSP), or application-specific integrated circuit (ASIC), suitable for embedded systems or specific functional scenarios. These processors all have specific instruction set architectures, capable of parsing and executing program instructions stored in computer-readable storage media.

[0191] Implementing a feedback processing method based on source weights refers to the entire process by which a processor, through executing a computer program and following preset logic and steps, dynamically manages the weights of feedback sources, receives feedback, calculates effective weights, accumulates weights, and ultimately adjusts the behavior of the target system. Specifically, the program includes an algorithm module for maintaining dynamic relationship weights with the target system, an interface module for receiving feedback to the target system, a mathematical model module for calculating the effective weights of the feedback based on the relationship weights, a database operation or memory management module for accumulating the effective weights, and a control module for adjusting the behavior of the target system when the accumulated weights meet predetermined conditions. Through the collaborative work of these modules, the abstract method can be run in a real-world computing environment.

[0192] This application provides a computer-readable storage medium storing a computer program, which is then executed by a processor, to implement a source-weight-based feedback processing method. This design makes the originally abstract feedback processing method concrete and operational. Specifically, the computer-readable storage medium, as a physical carrier, carries all the instructions and data for implementing the method, ensuring the persistent storage and distributability of the method logic. When the storage medium is loaded into a computer system, the processor can read and parse these computer program instructions. The processor executes a series of steps according to the preset logical order in the program, including maintaining dynamic relationship weights, receiving feedback, calculating effective weights, accumulating effective weights, and adjusting the target system behavior based on the accumulated weights. The ingenuity of this working principle lies in decoupling the complex feedback processing logic from the hardware level and implementing it in software, greatly improving the flexibility and maintainability of the method. In this way, the source-weight-based feedback processing method is no longer just a theoretical concept, but can be deployed on various computing devices, such as intelligent agents, industrial control systems, vehicle systems, IoT platforms, recommendation systems, search engines, or online service platforms. When the processor executes the program, it can dynamically maintain dynamic relationship weights between each feedback source and the target system, and update these weights based on factors such as interaction frequency and historical accuracy. Received feedback is used to calculate effective weights based on these dynamic weights, and these weights are accumulated. When the accumulated weights meet predetermined conditions, the processor triggers corresponding behavioral adjustments, such as adjusting the target system's response strategy or updating relationship data. This overall operating mechanism enables the feedback processing method proposed in this application to overcome the problems of homogeneous feedback, static weights, single processing methods, and lack of learning ability in traditional feedback processing. It achieves the goals of anti-interference, self-adaptation, and continuous optimization, and can play a role in practical applications in an efficient and standardized manner.

[0193] As a specific implementation, one can envision a smart assistant system where the core feedback processing logic is coded as a Python program. This Python program, after being compiled or interpreted, is stored on a solid-state drive (SSD), serving as a computer-readable storage medium. The main control unit of the smart assistant system, such as an ARM-based embedded processor, is responsible for executing this program. When the smart assistant starts, the processor loads the Python program from the SSD into its memory. After the program begins running, the processor, according to program instructions, first initializes and maintains a dynamic relationship weight table stored in memory or persistent storage. This table records the relationship weights between different users (feedback sources) and the smart assistant (target system). When a user provides feedback to the smart assistant via voice or text, such as expressing satisfaction or dissatisfaction with a response, the smart assistant system captures these feedback events through its feedback receiving module. The processor executes the weight calculation logic in the program, calculating the effective weight of the feedback based on the current user's relationship weight. Subsequently, the processor accumulates this effective weight into a corresponding accumulated weight variable, which can be distinguished by user and feedback type. When the accumulated negative feedback weights exceed a preset threshold, the processor executes the behavior adjustment logic in the program. For example, it adjusts the intelligent assistant's response strategy to the user, making it more cautious or providing more personalized services in subsequent interactions. Throughout the process, the processor continuously executes program instructions to ensure that each step of the feedback processing method is executed accurately and efficiently.

[0194] By providing a computer-readable storage medium, a computer program, and a processor to execute the program, this application effectively solves the problem that the lack of an effective executable form when deploying source weight-based feedback processing methods to computer systems leads to complex implementation, difficulty in updating, and affects the widespread application and efficiency of the method. Specifically, this implementation method allows complex feedback processing logic to exist in a standardized software form, greatly simplifying the deployment and updating process of the method and avoiding the high costs and inefficiencies of traditional hardware implementations or manual intervention. The computer-readable storage medium, as the physical carrier, ensures the portability and standardization of the method, enabling it to run seamlessly on different types of computing devices. The processor executing the computer program automates the feedback processing logic, thereby improving processing efficiency and accuracy. More importantly, through this executable form, the dynamic weight, cumulative triggering, and closed-loop learning feedback processing method proposed in this application can fully leverage its anti-interference, adaptability, and continuous optimization capabilities in practical applications, enabling the target system to respond to external feedback more robustly and intelligently, thereby significantly improving the overall system performance and user experience.

[0195] Other application scenarios The following simplified embodiments illustrate the application of the present invention in other fields. Specific implementations of these embodiments can be found in the examples described in the detailed embodiments above, and will not be repeated here.

[0196] Simplified Example 1: IoT Sensor Data Fusion In IoT systems, different sensors generate data with varying accuracy. The system dynamically adjusts the weights of these sensors based on their historical data accuracy. When a sensor reports data (which can be considered positive or negative feedback), the system accumulates the weighted data value. Only when the accumulated value exceeds a threshold is a corresponding action (such as an alarm) triggered. Data from low-accuracy sensors needs to be reported multiple times to trigger an alarm, thus avoiding false alarms.

[0197] Simplified Implementation Example 2: Driver Command Priority in Vehicle Systems In intelligent vehicles, different drivers have different driving habits and safety preferences. The system dynamically adjusts the weight of each driver's commands based on their historical driving records (such as the number of times they braked suddenly and their traffic violations). When a driver issues a command (such as accelerating or changing lanes), the system accumulates the command weight, and only executes it when it exceeds a threshold. For novice drivers, their commands require multiple confirmations or accumulate higher weights before execution, thus improving safety.

[0198] Simplified Example 3: Search Engine Ranking Optimization Search engines dynamically adjust their ranking algorithms based on user clicks and feedback on search results. Feedback from high-frequency users carries higher weight, ensuring search optimization better aligns with the needs of core users. The system updates user engagement metrics based on user dwell time and click behavior, achieving precise optimization.

[0199] Simplified Example 4: Content Recommendation on Social Media Social media platforms optimize their content recommendation strategies based on user interactions with content (likes, shares, reports) and the closeness of the user's relationship with the platform. Feedback from active creators carries higher weight to prevent malicious reports from interfering with the recommendation algorithm.

[0200] Simplified Example 5: Multi-Agent Collaborative Feedback Learning In a multi-agent system, there are multiple agents, including a master agent, child roles, and elder roles. Each agent forms a dynamic relationship of intimacy with other agents. For example, the master agent interacts frequently with child roles and collaborates to complete tasks, resulting in a high intimacy level; it interacts less with elder roles, resulting in a lower intimacy level. When a child role provides feedback (such as a like or dislike) on a decision made by the master agent, the system calculates the feedback weight based on the intimacy level between the child role and the master agent; the higher the intimacy level, the greater the impact of the feedback. Simultaneously, feedback between child roles also affects their relationships. For example, if two child roles complete a task together and then give each other positive feedback, their intimacy level increases, leading to smoother future collaboration. This embodiment demonstrates the wide application of this application in multi-agent scenarios, achieving personalized feedback learning between agents through dynamic relationship intimacy.

[0201] Simplified Example 6: Feedback Learning from External System API Calls When an agent calls an external weather API, the user experience may be affected by fluctuations in the API's service quality. The agent can learn from the API by analyzing the success rate, response time, and data accuracy of each call.

[0202] process: Feedback source identification: Weather API is used as the feedback source.

[0203] Relationship parameter calculation: The initial relationship parameter is 0.5. Each successful call increases the parameter by 0.01, a failed call decreases it by 0.02, and a response time exceeding a threshold decreases it by 0.01.

[0204] Feedback weight calculation: When the quality of API service deteriorates, the agent records negative feedback and calculates the weight based on the current relationship parameter (assumed to be 0.4), which is then accumulated into the negative feedback record of that API.

[0205] Behavioral adjustment: When negative feedback accumulates to a threshold, the agent reduces the frequency of calls to that API or switches to an alternative API. Simultaneously, the feedback event is stored in memory for future optimization.

[0206] Technical effect: The intelligent agent can adaptively select the optimal external service, improving the robustness of the system.

[0207] The above descriptions are merely embodiments of this application and are not intended to limit the scope of protection of this application. It should be noted that the term "user" in this application is a preferred embodiment of the feedback source, but not a limitation on the scope of protection. Those skilled in the art should understand that the feedback source can be any entity capable of interacting with the interactive system, such as a human user, other systems, devices, or applications. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

[0208] While this solution is applicable to local devices, it can also be deployed on cloud servers. In a cloud environment, larger-scale embedded models and vector databases can be used to further improve retrieval accuracy and capacity. Those skilled in the art should understand that the technical solution of this invention is not limited to a specific deployment environment, and any implementation based on the technical concept of this application should be considered to fall within the protection scope of this invention.

Claims

1. A feedback processing method based on source weight, characterized in that, Includes the following steps: Maintain dynamic relationship weights between each feedback source and the target system; Receive feedback to the target system; The effective weight of the feedback is calculated based on the relationship weight, wherein the effective weight increases as the relationship weight increases; Accumulate the effective weights; When the accumulated weights meet predetermined conditions, the behavior of the target system is adjusted.

2. The method of claim 1, wherein, The accumulation includes accumulation by feedback source and feedback type.

3. The method of claim 1, wherein, The feedback sources include, but are not limited to, at least one of the following: user, intelligent agent, external system, device, or internal module of the system.

4. The method of claim 1, wherein, The dynamic relationship weights are updated based on at least one of the following: interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, or corrective behavior.

5. The method of claim 4, wherein, The relationship weight update includes positive increments and negative decays, and is limited to a preset range.

6. The method of claim 1, wherein, The feedback includes at least one of positive feedback, negative feedback, and correction.

7. The method of claim 1, wherein, The predetermined conditions include the cumulative negative feedback weight exceeding a threshold.

8. The method of claim 1, wherein, The actions of adjusting the target system include, but are not limited to, at least one of adjusting the target system's response strategy, updating relational data, or reducing its confidence on relevant topics.

9. The method of claim 1, wherein, Also includes: The accumulated weights are decayed over time so that the impact of recent feedback is greater than that of long-term feedback.

10. The method of claim 9, wherein, The time decay adopts a preset decay model to periodically decay the accumulated weight.

11. The method of claim 1, wherein, It also includes storing feedback events and changes in relationship weights in a storage module for subsequent analysis and learning.

12. The method according to any one of claims 1 to 11, characterized in that, The target system includes, but is not limited to, intelligent agents, industrial control systems, vehicle systems, Internet of Things platforms, recommendation systems, search engines, or online service platforms.

13. The method according to any one of claims 1 to 12, characterized in that, As one implementation method, the method can be deployed on a local device or a cloud server.

14. A feedback processing system based on source weight, the system comprising: include: The relationship weight maintenance module is used to maintain the dynamic relationship weight between each feedback source and the target system; A feedback receiving module is used to receive feedback to the target system; The weight calculation module is used to calculate the effective weight of the feedback based on the relationship weight, wherein the effective weight increases as the relationship weight increases; The accumulation module is used to accumulate the effective weights; The triggering module is used to adjust the behavior of the target system when the accumulated weights meet predetermined conditions.

15. The system of claim 14, wherein, The accumulation module is configured to accumulate the effective weights according to the feedback source and feedback type, respectively.

16. The system of claim 14, wherein, It also includes a storage module for storing feedback events and changes in relationship weights.

17. The system of claim 14, wherein, It also includes a decay module for decaying the accumulated weights over time.

18. The system of claim 14, wherein, The relationship weight maintenance module is configured to update relationship weights based on at least one of the following: interaction frequency, historical accuracy, collaboration records, role permissions, sharing degree, or correction behavior.

19. The system of claim 14, wherein, The triggering module is configured to trigger an adjustment when the accumulated negative feedback weight exceeds a threshold.

20. The system of claim 14, wherein, The adjustments include, but are not limited to, at least one of the following: adjusting the target system's response strategy, updating relational data, or reducing its confidence level on relevant topics.

21. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any one of claims 1 to 13.

Images