Information processing method and apparatus, computing device

Abstract

Embodiments of the present application provide an information processing method and device, and a computing device. The method comprises: detecting a system access request initiated by a target user for a target processing system; in response to the system access request, determining decision results respectively generated by a plurality of online decision models for the target processing system; determining target recommendation information based on the decision results respectively corresponding to the plurality of online decision models; and outputting the target recommendation information for the target user. Embodiments of the present application improve recommendation accuracy and effectiveness.

Description

Technical Field

[0001] This application relates to the field of electronic equipment technology, and in particular to an information processing method and apparatus, and a computing device. Background Technology

[0002] An online decision-making system is an access platform that generates recommended content based on an online decision-making model, presents this content to users online, and continuously updates the recommended content by rapidly obtaining user feedback and adjusting the online decision-making model accordingly. Examples include online course systems, online trading systems, social networking systems, and online financial systems. Online course systems can recommend courses, online trading systems can recommend products, social networking systems can recommend social content, and online financial systems can recommend financial products.

[0003] In existing technologies, online decision-making models achieve convergence of the online decision-making algorithm by balancing exploration and exploitation to obtain the recommended content that generates the highest feedback. Online decision-making algorithms can learn to acquire an online decision-making model. During the learning process, the administrator can determine the online learning algorithm, set its hyperparameters based on experience for modeling, and then train it using training data to obtain the model parameters. After the online decision-making model is determined, its model parameters can be continuously updated based on user feedback on the recommended content.

[0004] However, online decision-making algorithms are usually selected based on human experience. When human experience is actually lacking, the algorithm settings may not be accurate enough, resulting in poor decision-making performance of online decision-making models and low effectiveness of recommended content. Summary of the Invention

[0005] In view of this, embodiments of this application provide an information processing method and apparatus, and a computing device, to solve the technical problem that online decision-making algorithms and their hyperparameters are usually set manually in the prior art, resulting in low recommendation effectiveness.

[0006] In a first aspect, embodiments of this application provide an information processing method, including:

[0007] Detect system access requests initiated by the target user against the target processing system;

[0008] In response to the system access request, determine the decision results generated by multiple online decision models in the target processing system;

[0009] Based on the decision results corresponding to the multiple online decision models, target recommendation information is determined.

[0010] Output the target recommendation information to the target user.

[0011] Secondly, embodiments of this application provide an information processing method, including:

[0012] Detect system access requests initiated by target users against the online transaction system;

[0013] In response to the system access request, determine the decision results generated by multiple online decision models in the online transaction system;

[0014] Based on the decision results corresponding to the multiple online decision models, the target recommendation information for the recommended transaction object is determined.

[0015] Output the target recommendation information of the recommended transaction object to the target user.

[0016] Thirdly, embodiments of this application provide an information processing method, including:

[0017] Detect system access requests initiated by target users for the content recommendation system;

[0018] In response to the system access request, determine the decision results generated by multiple online decision models in the content recommendation system;

[0019] Based on the decision results corresponding to the multiple online decision models, the target recommended content is determined.

[0020] Output the target recommended content to the target user.

[0021] Fourthly, embodiments of this application provide an information processing method, including:

[0022] In response to a call request to the adversarial decision-making interface, determine the information processing resources corresponding to the adversarial decision-making interface;

[0023] The following steps are performed using the information processing resources corresponding to the adversarial decision-making interface:

[0024] The adversarial decision interface is used to detect system access requests initiated by the target user against the target processing system;

[0025] In response to the system access request, determine the decision results generated by multiple online decision models in the target processing system;

[0026] Based on the decision results corresponding to the multiple online decision models, target recommendation information is determined.

[0027] The target recommendation information is output to the target user using the adversarial decision interface.

[0028] Fifthly, embodiments of this application provide an information processing apparatus, including:

[0029] The request detection module is used to detect system access requests initiated by target users for the target processing system;

[0030] The request and response module is used to respond to the system access request and determine the decision results generated by the multiple online decision models in the target processing system.

[0031] The information determination module is used to determine target recommendation information based on the decision results corresponding to the multiple online decision models.

[0032] The information output module is used to output the target recommendation information to the target user.

[0033] In a sixth aspect, embodiments of this application provide a computing device, including: a storage component and a processing component; the storage component is used to store one or more computer instructions; the one or more computer instructions are invoked by the processing component;

[0034] The processing component is used for:

[0035] The system detects system access requests initiated by target users for the target processing system; in response to the system access requests, it determines the decision results generated by multiple online decision models in the target processing system; based on the decision results corresponding to the multiple online decision models, it determines target recommendation information; and outputs the target recommendation information to the target user.

[0036] In this embodiment, upon receiving a system access request from a target user for a target processing system, the system can respond to the request by identifying multiple online decision models within the target processing system. Then, based on the recommended target of the target processing system, the system obtains the decision results corresponding to each of the multiple online decision models. Furthermore, by fusing these multiple decision results, target recommendation information can be obtained. This target recommendation information can then be output to the target user. By simultaneously making decisions using multiple online decision models, and by integrating the decision results of these models for recommendation, a comprehensive recommendation integrating multiple online decision models is obtained. This achieves multi-model comprehensive recommendation, effectively considering the decision results of multiple models, increasing the effectiveness and reliability of the target recommendation information, and improving recommendation efficiency. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart illustrating one embodiment of an information processing method provided in this application;

[0039] Figure 2 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0040] Figure 3 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0041] Figure 4 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0042] Figure 5 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0043] Figure 6 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0044] Figure 7 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0045] Figure 8 An application example diagram of an information processing method provided in this application embodiment;

[0046] Figure 9 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0047] Figure 10 A flowchart illustrating yet another embodiment of an information processing method provided in this application;

[0048] Figure 11 A schematic diagram of the structure of an embodiment of an information processing device provided in this application;

[0049] Figure 12 This is a schematic diagram of one embodiment of a computing device provided in this application. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one.

[0052] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0053] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to identification.” Similarly, depending on the context, the phrases “if determination” or “if identification (of the condition or event of the statement)” can be interpreted as “when determination” or “in response to determination” or “when identification (of the condition or event of the statement)” or “in response to identification (of the condition or event of the statement).”

[0054] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0055] The technical solution of this application embodiment can be applied to online decision-making scenarios. By employing multiple online decision-making models to make online recommendations simultaneously and fusing multiple decision results, effective recommendations can be achieved and recommendation efficiency can be improved.

[0056] In existing technologies, online decision-making systems often involve choices made by management during the selection of online decision-making algorithms, the construction of online decision-making models, and even the subsequent training of these models. Therefore, online decision-making algorithms are used to generate recommended content, which is then displayed to users to achieve the desired recommendation. User feedback is then gathered, and the online decision-making algorithm is updated based on this feedback. However, in this process, model selection and user models are determined by expert users. If these expert users lack experience, it may lead to poor model accuracy and unsatisfactory results.

[0057] In this embodiment, when a system access request initiated by a target user for a target processing system is received, the system can respond to the access request by determining the decision results corresponding to multiple online decision models of the target processing system; then, target recommendation information can be obtained based on the multiple decision results. This target recommendation information can then be output to the target user. By making decisions simultaneously through multiple online decision models, and by integrating the decision results of multiple online decision models for recommendation, a comprehensive recommendation based on multiple models is achieved. This effectively considers the decision results of multiple models, increases the effectiveness and reliability of the target recommendation information, and improves recommendation efficiency.

[0058] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0059] like Figure 1 The diagram shown is a flowchart of one embodiment of an information processing method provided in this application. The method may include the following steps:

[0060] 101: Receive system access requests initiated by the target user for the target processing system.

[0061] The information processing method provided in this application can be applied to computing devices, such as computers, servers, cloud servers, supercomputers, laptops, tablets, etc. This application does not impose excessive limitations on the specific type of computing device.

[0062] Optionally, the target processing system can be an online platform that recommends products or content to users. Typically, it can select K items or products from N candidate items or products to recommend to the user, where N is greater than K. Target processing systems can include, for example, online transaction systems, social network systems, e-commerce systems, online education systems, etc. This application embodiment does not impose excessive limitations on the specific type of target processing system.

[0063] System access requests can be initiated by the target user. The client can detect the access operation initiated by the target user against the target processing system, and in response to the access operation, generate a system access request and send the system access request to the computing device.

[0064] 102: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0065] Online decision-making models enable target processing systems to make online decisions and generate corresponding outcomes. These models can include machine learning models. Based on user-initiated system access requests, online decision-making models can make decisions for users and generate feedback. For example, in e-commerce applications, online decision-making models can select objects from multiple candidates to recommend to the user; the selected objects become the decision results.

[0066] An online decision model can be the output of an online decision-making algorithm running on training data. An online decision-making model can be obtained by running the algorithm on the training data. An online decision-making algorithm can be a decision-making method or process, and the algorithm can build a model based on learning from the training data. For example, in online decision-making algorithms, UCB (Upper Confidence Bound Algorithm) and TS (Thompson Sampling) can be used to generate UCB and TS models by running the algorithms on the training data. Both the generated UCB and TS models can serve as online decision-making models. The training data for an online decision-making algorithm can be real-time feedback data generated from user interactions with recommended content.

[0067] Multiple online decision-making models can include online decision-making models with the same algorithm but different model hyperparameters. Taking the UCB algorithm as an example, three hyperparameters can be set, and online decision-making models can be built using the three hyperparameters respectively. After training the constructed hyperparameters, three UCB models can be obtained.

[0068] Optionally, multiple online decision models can be determined in the following way: in response to a system access request, multiple online decision algorithms in the target processing system are determined; for any online decision algorithm, at least one hyperparameter of the online decision algorithm is obtained, and online decision models constructed using the at least one hyperparameter are trained using training data to obtain at least one online decision model corresponding to the online decision algorithm, and then all online decision models corresponding to the at least one online decision algorithm are obtained to form multiple online decision models.

[0069] 103: Determine target recommendation information based on the decision results corresponding to multiple online decision models.

[0070] Multiple online decision-making models can generate decision results independently. Typically, the decision results are primarily based on the recommendation targets obtained by the target processing system. For example, in an e-commerce system, the recommendation target is to recommend products to users; the online decision-making model can recommend products online, and the target recommendation information can be the target product information. In a social networking system, the recommendation target can be the content recommended to users; the online decision-making model can recommend trending news, trending events, etc., and the target recommendation content can be the target recommended content.

[0071] The target recommendation information integrates multiple decision results, combining the decision results of multiple online decision models to achieve comprehensive recommendation from multiple online decision models. This avoids the simplistic performance limitations of using a single online decision model, enabling effective recommendation of decision results and improving recommendation performance.

[0072] 104: Output target recommendation information for the target user.

[0073] Optionally, the target recommendation information for the target user may specifically include: sending the target recommendation information to the target user's client so that the client can output the target recommendation information.

[0074] In this embodiment, after receiving a system access request from a target user for the target processing system, the system can respond to the request by determining multiple online decision models of the target processing system. By acquiring the decision results corresponding to each of the multiple online decision models and fusing these results, target recommendation information is obtained and then output to the target user. By simultaneously making decisions through multiple online decision models and integrating the results, target recommendation information that incorporates the individual decision results of each online decision model is obtained. This achieves comprehensive recommendation across multiple models, effectively considering the decision results of multiple online decision models, increasing the effectiveness and reliability of the target recommendation information, and improving recommendation efficiency.

[0075] To obtain accurate decision results, as an example, acquiring the decision results generated by multiple online decision models includes:

[0076] Based on the recommended targets of the target processing system, the decision results generated by multiple online decision models are obtained.

[0077] Optionally, the recommendation objective may include constraints that affect the computational results of the online decision-making model. These constraints regulate the decision-making results of the online decision-making model, requiring the results to meet these constraints. For example, in an e-commerce system, the recommendation objective could be products with a click-through rate exceeding a predetermined threshold; in a social networking system, it could be trending events with a certain number of clicks; and in an online education system, it could be the identity information of teachers with high evaluation scores.

[0078] The decision outcome generated by any online decision-making model may include: predicting the usage effects of multiple candidate objects, obtaining recommendation data corresponding to each candidate object, and determining at least one recommended object that satisfies the constraints from the recommendation data corresponding to each candidate object. Each online decision-making model generates at least one recommended object, wherein any recommended object can be recommended by one or more online decision-making models.

[0079] After determining at least one recommended object generated by at least one online decision-making model, the target recommendation information is determined based on the decision results corresponding to multiple online decision-making models. Specifically, this may include: determining the recommendation data for any given recommended object across multiple online decision-making models; wherein, if any online decision-making model does not recommend the recommended object, the recommendation data for that object in that online decision-making model is 0. Based on the multiple recommendation data corresponding to any given recommended object, the average recommendation data for that object is calculated, thereby determining the average recommendation data for all recommended objects. Based on the average recommendation data for all recommended objects, at least one target object is selected from all recommended objects, thereby determining the target recommendation information composed of at least one target object.

[0080] Multiple online decision-making models can be approached from two perspectives: algorithm selection and parameter selection. For example... Figure 2 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include the following steps:

[0081] 201: Receive system access requests initiated by the target user for the target processing system.

[0082] Some steps in the embodiments of this application are similar to Figure 1 Some steps are the same in the illustrated embodiments, and for the sake of brevity, they will not be repeated here.

[0083] 202: Select at least one objective decision algorithm from multiple online decision algorithms.

[0084] At least one objective decision algorithm can be selected from multiple online decision algorithms.

[0085] For ease of understanding, let's take as an example several online decision-making algorithms that may include TS (Thompson Sampling), UCB (Upper Confidence Bound), LinUCB (Linear Upper Confidence Bound), LinTS (Linear Thompson Sampling), and Exp3 (Exponential-weight algorithm for exploration and Exploitation). We can then select LinTS or LinUCB as the target decision-making algorithm from among these algorithms.

[0086] 203: Obtain at least one online decision model corresponding to any target decision algorithm, so as to obtain multiple online decision models composed of at least one online decision model of each of the at least one target decision algorithms.

[0087] 204: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0088] Optionally, the target processing system can provide multiple candidate objects for the online decision-making model to make a decision. Any online decision-making model can predict the probability or score of recommending multiple candidate objects to obtain a decision result composed of recommendation data corresponding to each candidate object. Recommendation data can include recommendation probabilities or recommendation scores. The size of the recommendation data indicates the likelihood of a candidate object being recommended; the larger the recommendation data, the higher the probability of the candidate object being recommended, and the smaller the recommendation data, the lower the probability of the candidate object being recommended.

[0089] 205: Determine target recommendation information based on the decision results corresponding to multiple online decision models.

[0090] The target recommendation information can be determined by the decision results of multiple online decision models. By combining the decision results of multiple online decision models, the target recommendation information can be more accurate and the recommendation efficiency can be higher.

[0091] In object recommendation scenarios, the decision results of any online decision model can include recommendation data predicted for multiple candidate objects. The steps for determining target recommendation information can include: obtaining recommendation data for each candidate object across multiple online decision models to determine the average recommendation data for that candidate object, thereby determining the average recommendation data for each of the multiple candidate objects; selecting K target objects from the multiple candidate objects based on their average recommendation data; and generating target recommendation information using the object information of each of the K target objects.

[0092] Optionally, the selection step of K target objects may include: selecting K target objects from multiple candidate objects in descending order of average recommendation data.

[0093] Optionally, the selection step of K target objects may include: determining the historical recommendation data corresponding to multiple candidate objects respectively; calculating the recommendation offset data based on the average recommendation data and historical recommendation data corresponding to any candidate object to obtain the recommendation offset data corresponding to multiple candidate objects respectively; and selecting K target recommendation objects whose recommendation offset data satisfies a preset offset threshold from multiple candidate objects according to the recommendation offset data corresponding to multiple candidate objects respectively.

[0094] The number of candidate objects is N, and K is a positive integer less than or equal to N and greater than or equal to 1.

[0095] 206: Output target recommendation information to the target user.

[0096] In this embodiment, when a system access request initiated by a target user for a target processing system is received, the system can respond to the access request by selecting at least one target decision algorithm from multiple online decision algorithms to obtain at least one online decision model corresponding to any one target decision algorithm. After initially selecting the algorithm, a decision model can be built based on the algorithm to obtain multiple online decision models, thus achieving accurate screening of online decision models. After determining multiple online decision models, the decision results corresponding to each of the multiple online decision models can be obtained, and the multiple decision results can be fused to obtain target recommendation information. The target recommendation information incorporates the decision results of multiple online recommendation models. By making decisions simultaneously through multiple online decision models, the recommendation is made by comprehensively considering the decision results of multiple online decision models, thereby obtaining a comprehensive recommendation that incorporates multiple online decision models. This effectively considers the decision results of multiple online decision models, increases the effectiveness and reliability of target recommendation information, and improves recommendation efficiency.

[0097] In one possible design, the system administrator can select the decision-making algorithm. Selecting at least one target decision algorithm from multiple online decision-making algorithms may include:

[0098] It provides multiple online decision-making algorithms for system management users.

[0099] Obtain at least one target decision algorithm selected by the system management user from multiple online decision algorithms.

[0100] By interacting with system administrators to select algorithms, effective algorithm selection can be achieved, making the algorithms more compatible with user habits and providing a more personalized algorithm selection service.

[0101] Online decision-making models can be determined through a combination of algorithm and hyperparameter selection. As one possible implementation, at least one online decision-making model corresponding to any objective decision-making algorithm can be determined in the following way:

[0102] Determine the hyperparameter setting rules for any objective decision-making algorithm;

[0103] Based on the hyperparameter setting rules, at least one hyperparameter of the target decision algorithm is determined;

[0104] Obtain the online decision models corresponding to the at least one hyperparameter, so as to obtain at least one online decision model corresponding to the target decision algorithm.

[0105] Each objective decision-making algorithm can correspond to a hyperparameter setting rule. This rule is used to generate at least one hyperparameter for the objective decision-making algorithm, and then at least one hyperparameter is used to construct a corresponding online decision-making model. Initially, the model parameters are unknown; training parameters are needed to obtain the model parameters for at least one online decision-making model.

[0106] For ease of understanding, let's take LinTS and LinUCB as the target decision algorithm as an example. Assume that LinUCB has a hyperparameter alpha, and the parameter value of alpha is in the range of [0,1]. Assume that the value of alpha is selected as 0.1, 0.5, 0.8 and 1.0 respectively. The resulting online decision models are: LinTS model, the first LinUCB model corresponding to LinUCB and alpha=0.1, the second LinUCB model corresponding to LinUCB and alpha=0.5, the third LinUCB model corresponding to LinUCB and alpha=0.8, and the fourth LinUCB model corresponding to LinUCB and alpha=1.0.

[0107] Optionally, the steps for obtaining the online decision-making models corresponding to at least one hyperparameter may include: constructing online decision-making models of the target decision-making algorithm corresponding to at least one hyperparameter, thereby obtaining at least one online decision-making model with unknown parameters; and training the models of the at least one online decision-making model using training data. The training process for the model parameters of any online decision-making model can refer to existing online decision-making model training methods, and will not be elaborated upon here.

[0108] The system characteristics of the target processing system can be features related to system properties. These system characteristics can serve as selection criteria for decision-making algorithms, leading to the acquisition of algorithms with higher correlation to system characteristics and thus achieving effective algorithm selection. As one possible implementation, selecting at least one target decision-making algorithm from multiple online decision-making algorithms may include:

[0109] Determine the system characteristics of the target processing system.

[0110] Based on system characteristic information, at least one target decision algorithm is selected from multiple online decision algorithms.

[0111] Optionally, system feature information may include: system function information, recommendation targets, relevant historical data, system version, system type and / or operating status information, etc. Relevant historical information may include: historical system usage information, historical recommendation information, historical operations performed by users on historical recommendation information, and historical operation data corresponding to those operations. When selecting at least one target decision algorithm from multiple online decision algorithms, the selection can be based on the system's feature information, which is related to the system's actual operating conditions. For example, the historical recommendation effect of the system can be judged by the historical operations performed by users on historical recommendation information. Therefore, if an online decision algorithm with a high historical recommendation effect obtains a large amount of recommendation data, that online decision algorithm can be used as the target decision algorithm; conversely, if another online decision algorithm obtains a small amount of recommendation data, it can be excluded from the target decision algorithm selection.

[0112] In one possible design, selecting at least one target decision algorithm from multiple online decision algorithms based on system characteristic information may include:

[0113] The first matching degree between multiple online decision-making algorithms and system feature information is determined respectively.

[0114] Based on the first matching degree corresponding to each of the multiple online decision-making algorithms, at least one target decision-making algorithm is selected from the multiple online decision-making algorithms.

[0115] The higher the first degree of matching between any online decision-making algorithm and the system's feature information, the higher the accuracy of the decision result of the online decision-making algorithm on the target processing system; the lower the first degree of matching, the lower the accuracy of the decision result of the online decision-making algorithm on the target processing system.

[0116] The initial matching degree between any online decision-making algorithm and system feature information can be obtained by scoring the matching scores between the online decision-making algorithm and the system feature information. For example, when the system feature information includes at least one system feature, the matching score between any online decision-making algorithm and at least one system feature can be estimated to obtain the matching score of each online decision-making algorithm and at least one system feature. Then, the matching scores of each online decision-making algorithm and at least one system feature can be summed or averaged to obtain the final matching score between the online decision-making algorithm and the system feature information. In practical applications, the matching relationships between any system feature and multiple online decision-making algorithms can be pre-stored. These matching relationships can be, for example, matching data or matching intervals. When at least one system feature of any target processing system is obtained, the matching relationships between any online decision-making algorithm and at least one system feature can be queried to estimate the matching score and obtain the final matching score between the online decision-making algorithm and at least one system feature.

[0117] The parameter setting rules for the target decision algorithm can be determined based on the algorithm's historical usage experience. In some embodiments, the model parameters of the online decision algorithm can satisfy a certain parameter value range, thereby ensuring that the obtained model parameters are within the required range. During the operation of the target processing system, parameters of multiple online decision algorithms can be set directly.

[0118] As another embodiment, determining at least one model parameter corresponding to the target decision algorithm based on the parameter setting rules of any target decision algorithm, so as to obtain at least one online decision model corresponding to the target decision algorithm, may include:

[0119] Based on the parameter value range corresponding to any target decision algorithm, at least one model parameter of the target decision algorithm is determined using a preset parameter selection rule, so as to obtain at least one online decision model corresponding to the target decision algorithm.

[0120] Optionally, at least one online decision model is set with at least one model parameter corresponding to any target decision algorithm to obtain at least one online decision model corresponding to the target decision algorithm. Performing at least one model sampling within the parameter selection range of the target decision algorithm according to the parameter selection rules ensures that the model parameters meet the parameter value range, thus obtaining accurate and effective model parameters.

[0121] The target processing system may include an online processing system that makes decision recommendations during system use. During the operation of the target processing system, system decisions can be initiated according to usage requirements, and parameters of the online decision-making algorithm can be set each time the system learning is initiated.

[0122] Furthermore, since the target processing system can read previously saved historical model parameters corresponding to multiple target decision algorithms as new model parameters during operation, parameter updates can be achieved. As another embodiment, determining at least one model parameter corresponding to a target decision algorithm based on the parameter setting rules of any given target decision algorithm to obtain at least one online decision model for that target decision algorithm may include: reading at least one historical model parameter corresponding to each target decision algorithm; determining the target historical parameter with the largest historical recommendation effect data among the at least one historical model parameter based on the historical recommendation effects corresponding to each of the at least one historical model parameter; re-sampling the target decision algorithm at least once based on the target historical parameter to obtain at least one model parameter corresponding to that target decision algorithm, thereby obtaining at least one online decision model for that target decision algorithm. Using the model parameter with the highest recommendation effect as the basis for parameter selection can improve the efficiency and accuracy of parameter selection.

[0123] In addition to system feature information, relevant user feature information, such as user identity information, interests and hobbies, historical collections, historical purchases, and historical browsing history, can also be used as a basis for algorithm selection. As another embodiment, selecting at least one target decision algorithm from multiple online decision algorithms may include:

[0124] Determine the user characteristics information of the target users;

[0125] Based on user characteristic information, at least one target decision algorithm is selected from multiple online decision algorithms.

[0126] When selecting at least one target decision algorithm from multiple online decision algorithms, the selection can be based on the target user's characteristic information, which is related to the target user's actual usage of the target processing system. For example, the target user's historical operations on historical recommendation information can be used to determine the historical recommendation effect of the target processing system. Thus, for objects with high historical recommendation effects, if a certain online decision algorithm obtains a large amount of recommendation data, that online decision algorithm can be selected as the target decision algorithm, while if another online decision algorithm obtains a small amount of recommendation data, it can be excluded from the target decision algorithm selection.

[0127] As one possible implementation, selecting at least one target decision algorithm from multiple online decision algorithms based on user characteristic information may include:

[0128] Determine the second degree of matching between multiple online decision-making algorithms and user feature information;

[0129] Based on the second matching degree corresponding to each of the multiple online decision-making algorithms, at least one target decision-making algorithm is selected from the multiple online decision-making algorithms.

[0130] The higher the second degree of matching between any online decision-making algorithm and user feature information, the higher the accuracy of the decision result of the online decision-making algorithm on the target processing system; the lower the second degree of matching, the lower the accuracy of the decision result of the online decision-making algorithm on the target processing system.

[0131] The second matching degree between any online decision-making algorithm and user feature information can be obtained by scoring the matching scores between the online decision-making algorithm and the user feature information. For example, when the user feature information includes at least one user feature, the matching score between any online decision-making algorithm and at least one user feature can be estimated to obtain the matching score of each online decision-making algorithm and at least one user feature. Then, the scores are summed or averaged to obtain the final matching score between the online decision-making algorithm and the user feature information. In practical applications, the matching relationships between any user feature and multiple online decision-making algorithms can be pre-stored. These matching relationships can be, for example, matching data or matching intervals. When at least one user feature of any target processing system is obtained, the matching relationships between any online decision-making algorithm and at least one user feature can be queried to estimate the matching score and obtain the second matching score between the online decision-making algorithm and at least one user feature.

[0132] In one possible design, user characteristics and system characteristics can be considered together to select the algorithm. In this case, selecting at least one target decision algorithm from multiple online decision algorithms can include:

[0133] Determine the user characteristics of the target user and the system characteristics of the target processing system;

[0134] Determine the first degree of matching between multiple online decision-making algorithms and system feature information;

[0135] Determine the second degree of matching between multiple online decision-making algorithms and user feature information;

[0136] Based on the first matching degree and the second matching degree corresponding to multiple online decision-making algorithms, at least one target decision-making algorithm is selected from multiple online decision-making algorithms.

[0137] The specific methods for determining the first and second matching degrees are the same as those in the aforementioned embodiments and will not be repeated here. Obtaining at least one target decision algorithm from multiple online decision algorithms based on their respective first and second matching degrees may include: determining the target matching degree of the target decision algorithm based on the first and second matching degrees of any given online decision algorithm; and selecting at least one target decision algorithm from the multiple online decision algorithms based on their target matching degrees. The target matching degree of any online decision algorithm can be obtained by calculating the mean, weighted average, and / or variance and covariance of the first and second matching degrees.

[0138] Optionally, when selecting at least one target decision algorithm from multiple online decision algorithms based on matching degree, the matching degrees corresponding to the multiple online decision algorithms can be sorted in descending order, and the top K target decision algorithms can be selected from the sorted online decision algorithms. The matching degree can be, for example, the first matching degree, the second matching degree, or the target matching degree.

[0139] like Figure 3 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include the following steps:

[0140] 301: Detects system access requests initiated by the target user against the target processing system.

[0141] 302: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0142] 303: The decision results corresponding to multiple online decision-making models are fused together to obtain the target decision result.

[0143] The target decision result integrates the decision results corresponding to multiple online decision models. Compared with the decision result obtained by a single online decision model in a single decision, the decision content is more comprehensive and the decision effectiveness is higher.

[0144] 304: Based on the target decision results, determine the target recommendation information.

[0145] 305: Output target recommendation information for the target user.

[0146] It should be noted that, in the embodiments of this application, the following applies: Figure 1 Some steps are the same in the illustrated embodiments, and will not be repeated here.

[0147] In this embodiment, when a system access request initiated by a target user for a target processing system is received, the system can respond to the system access request by determining the decision results corresponding to multiple online decision models. The target decision result is obtained by fusing the multiple decision results. Then, target recommendation information is obtained using the target decision result. The target decision result incorporates the decision results of multiple online recommendation models, and the target recommendation information obtained from the target decision result also integrates the decision results of multiple online decision models. Through simultaneous decision-making by multiple online decision models, comprehensive recommendation from multiple online decision models is achieved, effectively considering the decision results of multiple models, increasing the effectiveness and reliability of the target recommendation information, and improving recommendation efficiency.

[0148] As an example, fusing the decision results from multiple online decision-making models to obtain the target decision result may include:

[0149] Determine the decision weights for each of the multiple online decision-making models;

[0150] Based on the decision weights corresponding to multiple online decision-making models, the decision results corresponding to the multiple online decision-making models are weighted to obtain the target decision result.

[0151] When multiple decision results can be integrated to obtain the target decision result, a weighted calculation method is adopted. The multiple decision results are weighted according to their corresponding decision weights to obtain the target decision result, thereby achieving accurate integration of decision results and improving the accuracy of integration.

[0152] The target processing system can be an online system. When a target user accesses the system, it displays targeted recommendation information. The target user can then perform corresponding actions based on this information. For example, in e-commerce, the recommendation information might include details about the target product. After displaying this information, the target user can browse the recommended product and perform actions such as clicking, viewing, or purchasing.

[0153] In some embodiments, after outputting target recommendation information to the target user, the method may further include:

[0154] Obtain feedback actions taken by the target user in response to the target recommendation information;

[0155] Based on the feedback data corresponding to the feedback operation, update the model parameters of multiple online decision-making models to obtain multiple online decision-making models with updated parameters.

[0156] The process of obtaining the updated decision results corresponding to the multiple online decision models and returning them to determine the target recommendation information continues.

[0157] Optionally, updating the model parameters of multiple online decision-making models based on the feedback data corresponding to the feedback operation to obtain multiple online decision-making models with updated parameters may include: adding the feedback data corresponding to the feedback operation to the training data, and using the new training data to train the model parameters of multiple online decision-making models respectively to obtain multiple online decision-making models with updated parameters.

[0158] Feedback actions taken by target users in response to target recommendations can indicate the effectiveness of the recommendations. For example, the effectiveness of a target recommendation is high when the user browses and purchases a target recommended item, and low when the user does not browse or purchase any target recommended item. User feedback actions can be used to update the decision weights used in online decision-making models and / or result fusion processes.

[0159] In practical applications, user feedback actions are typically descriptive of user behavior. To represent user behavior digitized and update online decision-making models and weights based on these actions, action identifiers can be used to record user actions. For example, suppose user actions include clicking, browsing, and purchasing. Identifier H1 can be used to identify a click, H2 to identify a browse, and H3 to identify a purchase. The action data corresponding to a feedback action can then include the corresponding action identifier.

[0160] In this embodiment of the application, the system access request initiated by the target user is used as the feedback basis. In one possible design, the target user can initiate multiple system access requests to the target processing system. After outputting target recommendation information to the target user for any system access request initiated by the target user, the feedback operation performed by the target user on the target recommendation information can be obtained.

[0161] When a target user initiates multiple system access requests, multiple feedback operations can be obtained. In this case, based on the feedback data corresponding to the feedback operations, the model parameters of multiple online decision-making models are updated. The resulting online decision-making models can include: based on the operation identifiers corresponding to each feedback operation, counting the number of times the target user performs each feedback operation on the target recommended object; and updating the model parameters of each online decision-making model based on the number of operations corresponding to each feedback operation. The obtained feedback operations can include currently obtained feedback operations and historical feedback operations corresponding to historical recommended objects. For example, the number of operations corresponding to each feedback operation could be the number of times the same recommended object has been viewed or purchased during the current and historical access processes.

[0162] like Figure 4 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include the following steps:

[0163] 401: Detects system access requests initiated by the target user against the target processing system.

[0164] It should be noted that some steps in this embodiment are the same as some steps in the previous embodiment, and for the sake of brevity, they will not be repeated here.

[0165] 402: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0166] 403: Determine target recommendation information based on the decision results corresponding to multiple online decision models.

[0167] 404: Outputs target recommendation information to the target user.

[0168] 405: Obtain the feedback action taken by the target user in response to the target recommendation information.

[0169] 406: Based on the feedback data corresponding to the feedback operation, update the model parameters of multiple online decision-making models to obtain multiple online decision-making models with updated parameters.

[0170] 407: Retrieve the updated decision results for each of the multiple online decision models, and return to step 403 to continue execution.

[0171] In this embodiment, after detecting a system access request initiated by a target user for the target processing system, the system can respond to the access request and determine the decision results generated by multiple online decision models in the target processing system. Then, based on the decision results corresponding to the multiple online decision models, target recommendation information is determined. This target recommendation information integrates the decision results corresponding to multiple online decision models, resulting in more comprehensive decision content and effectively improved recommendation performance. After outputting the target recommendation information to the target user, the system can obtain the feedback operations performed by the target user in response to the target recommendation information. Based on the feedback data corresponding to these feedback operations, the model parameters of each of the multiple online decision models are updated, resulting in multiple online decision models with updated parameters. Obtaining the updated decision results corresponding to the multiple online decision models enables real-time updates of the multiple online decision models by utilizing user feedback operations. This strengthens the correlation between the decision results of the multiple online decision models and the user's actual operations, improving the operational correlation graph between the multiple online decision models and the target user. This allows the models to effectively learn from the user's operational directions, leading to more accurate decision results when using the multiple online decision models to make decisions for the target processing system again, thus achieving more accurate and efficient information recommendation.

[0172] Optionally, in practical applications, target feedback data can be stored in the form of operation logs. Operation logs are files generated by recording the relevant operations performed by the target user on the target recommendation information, such as logs corresponding to actions like clicking, browsing, and purchasing. Each operation can be identified using a different operation identifier, and operation logs are generated according to the log recording format based on the operation identifier. In this case, obtaining the feedback operations performed by the target user on the target recommendation information can include: parsing the operation logs performed by the target user on the target recommendation information to obtain the operation identifiers; determining the target feedback data corresponding to the feedback operations can include determining the target feedback data corresponding to the operation identifiers of the feedback operations.

[0173] After a target user views the target recommendation information, corresponding feedback operations can be performed on that information. These feedback operations represent the target user's response to the decision. Taking an e-commerce scenario as an example, if a product is recommended to a target user, and the user clicks and browses the product, it's confirmed that the recommendation has received feedback. If the user does not click or browse the product, it's confirmed that no feedback has been received. Feedback operations can include: "Received Feedback 1," "No Feedback 0." Since feedback operations are highly correlated with the target user, these user-executed feedback operations can be added as new training data to the online decision-making model's training process. This allows for updating the model parameters of multiple online decision-making models using the target feedback data, resulting in online decision-making models corresponding to the new parameters. The model parameters of any online decision-making model can be updated using the target feedback data to obtain the online decision-making model corresponding to the new parameters. After obtaining multiple new online decision-making models, the process can return to the step of responding to system access requests and determining the multiple online decision-making models of the target processing system to continue execution.

[0174] Optionally, updating the model parameters of multiple online decision-making models based on the target feedback data can include: adding the target feedback data to the training data, and then training each online decision-making model separately based on the updated training data to obtain the model parameters of each model. Assuming the original training data is the first training data, the target feedback data can be added to the first training data to obtain the second training data; the multiple online decision-making models can then be trained separately using the training data to obtain new model parameters for each model.

[0175] When fusing multiple decision outcomes, the decision weights corresponding to each outcome can be used to weight the results and obtain target recommendation information. However, the decision weights corresponding to the multiple outcomes can also be the system weights of the corresponding online decision-making systems.

[0176] In some embodiments, the decision results corresponding to multiple decision results can be determined by: determining the system weights corresponding to multiple online decision systems, and using the system weight of any online decision model as the decision weight corresponding to the decision result of that online decision model.

[0177] The system weights of any online decision-making model can be determined based on the historical feedback data corresponding to the online decision-making model. In one possible design, the number of feedbacks to recommended objects in the target recommendation information can be counted in the historical feedback data, and the system weight of the online decision-making model corresponding to the recommended object with a higher number of feedbacks can be determined to be higher than the system weight of the online decision-making model corresponding to the recommended object with a lower number of feedbacks.

[0178] During the operation of the target processing system, the system weights corresponding to multiple online decision models can be updated using user target feedback data. After the system weights are updated, the decision weights corresponding to multiple decision results are updated along with the system weights of the corresponding online decision systems.

[0179] like Figure 5 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include the following steps:

[0180] 501: Detects system access requests initiated by the target user against the target processing system.

[0181] 502: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0182] 503: Determine the decision weights for each of the multiple online decision models.

[0183] 504: Based on the decision weights corresponding to multiple online decision models, the decision results corresponding to the multiple online decision models are weighted and processed to obtain the target decision result.

[0184] 505: Output target recommendation information to the target user.

[0185] 506: Obtain the feedback action taken by the target user in response to the target recommendation information.

[0186] 507: Based on the feedback data corresponding to the feedback operation, update the decision weights corresponding to the multiple online decision models respectively, and return to step 504 to continue execution.

[0187] It should be noted that some steps in the embodiments of this application are different from those in the present application. Figure 1 Some steps are the same in the illustrated embodiments, and for the sake of brevity, they will not be repeated here.

[0188] Optionally, updating the decision weights of multiple online decision models based on the feedback data corresponding to the feedback operation may include: determining the recommendation effect of each of the multiple online decision models based on the feedback data corresponding to the feedback operation; determining the weight adjustment data of each of the multiple online decision models based on the recommendation effect of each of the multiple online decision models; updating the decision weights of each of the multiple online decision models based on the weight adjustment data of each of the multiple online decision models; and obtaining the updated decision weights of each of the multiple online decision models.

[0189] In one possible design, the weight adjustment data can include the adjustment direction and the adjustment step size. The recommendation performance of each online decision-making model can be data such as recommendation scores or actual click-through rates. When the recommendation performance of an online decision-making model is higher, the recommendation is more effective, and the decision weight of that model can be increased. Conversely, when the recommendation performance of an online decision-making model is lower, the recommendation is more ineffective, and the decision weight can be decreased. The increase or decrease in decision weight constitutes the adjustment direction, while the specific value of the increase or decrease constitutes the adjustment step size. The adjustment step size can be determined based on the differences in recommendation performance between different online decision-making models. For example, the recommendation performance of multiple online decision-making models can be normalized, and the normalized performance data can be compared with the original decision weights to determine the adjustment step size and direction. For instance, if the normalized performance data of an online decision-making model is 0.4 and the original decision weight is 0.1, the adjustment direction can be determined as an increase, and the adjustment step size can be determined based on the difference between 0.4 and 0.1. For example, taking three adjustments as an example, the adjustment step size for each adjustment could be 0.1. In this embodiment, when detecting a system access request initiated by a target user for a target processing system, the system can respond to the access request and determine the decision results generated by multiple online decision models in the target processing system. After determining the decision weights corresponding to the multiple online decision models, the decision results corresponding to the multiple online decision models can be weighted according to their respective decision weights to obtain the target decision result, and then target recommendation information can be output to the target user. After determining the target recommendation information, the target user can perform corresponding recommendation operations on the target recommendation information. Then, based on the feedback data corresponding to the user's feedback operations on the target recommendation information, the decision weights corresponding to the multiple online decision models are updated, realizing the update of decision weights. The updated weights are then used to update the target decision result again, realizing the effective update of parameters. Thus, when multiple online decision models are used again to make decisions for the target processing system, more accurate decision results can be obtained, thereby achieving more accurate and efficient information recommendation.

[0190] like Figure 6 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include:

[0191] 601: Detects system access requests initiated by the target user against the target processing system.

[0192] 602: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0193] 603: Determine the decision weights corresponding to the multiple online decision models.

[0194] 604: Based on the decision weights corresponding to multiple online decision models, the decision results corresponding to the multiple online decision models are weighted and processed to obtain the target decision result.

[0195] 605: Based on the target decision results, determine the target recommendation information.

[0196] 606: Output target recommendation information to the target user.

[0197] 607: Obtain the feedback actions taken by the target user in response to the target recommendation information;

[0198] 608: Based on the feedback data corresponding to the feedback operation, update the model parameters of multiple online decision-making models to obtain multiple online decision-making models with updated parameters.

[0199] 609: Obtain the decision results corresponding to the updated online decision models.

[0200] 610: Based on the feedback data corresponding to the feedback operation, update the decision weights corresponding to multiple online decision models respectively, and continue execution in step 604.

[0201] In this embodiment, after obtaining the target recommendation information as output by the target user, the feedback operations performed by the target user on the target recommendation information can be utilized. Then, based on the feedback data corresponding to the feedback operations, the model parameters of multiple online decision-making models are updated, and the decision weights corresponding to each of the multiple online decision-making models are updated simultaneously. The updated multiple online decision-making models are then used to re-make decisions, obtaining the decision results corresponding to each of the multiple online decision-making models. Subsequently, the decision results are weighted using the updated decision weights corresponding to each of the multiple online decision-making models to obtain the target decision result. This achieves real-time updates of the decision models and decision weights, and thus, when the multiple online decision-making models are used again to make decisions for the target processing system, more accurate decision results can be obtained, thereby achieving more accurate and efficient information recommendation.

[0202] To manage multiple online decision-making models more effectively, in one possible design, after updating the decision weights corresponding to the multiple decision outcomes, the method may further include:

[0203] The system administrator outputs the updated decision weights for each of the multiple online decision models.

[0204] To enable system administrators to adjust decision weights according to their management needs, as an example, after outputting the updated decision weights for multiple decision results to the system administrators, the system may further include:

[0205] The system detects weight adjustment operations triggered by the user on multiple decision weights and obtains the weight adjustment information input by the user.

[0206] Based on the weight adjustment information, multiple decision weights are updated to obtain updated decision weights. Then, based on the decision weights corresponding to the multiple online decision models, the decision results corresponding to the multiple online decision models are weighted and processed to obtain the target decision result for further execution.

[0207] In another possible design, after updating the model parameters of multiple online decision models based on the feedback data corresponding to the feedback operation, and obtaining the multiple online decision models with updated parameters, the method may further include:

[0208] Obtain the model parameters of each of the multiple online decision models that are candidates for updating.

[0209] For any given online decision-making model, determine the model convergence information corresponding to the updated model parameters of the online decision-making model, so as to obtain the model convergence information corresponding to multiple online decision-making models respectively.

[0210] Based on the model convergence information corresponding to multiple online decision-making models, the operational status information of the target processing system is generated.

[0211] Based on the operational status information, the target operation content of the target processing system is determined.

[0212] Perform target operations on the target processing system that match the target operation content.

[0213] After the model parameters of multiple online decision-making models are determined, the convergence degree of each model parameter within its respective online decision-making model can be determined based on the corresponding model parameters, thus obtaining model convergence information for each online decision-making model. This model convergence information can be used to determine the convergence of the online decision-making models and monitor the update results of the model parameters. During system operation, operational status information can be continuously generated, including not only the currently acquired operational status information but also multiple historical operational status information. Based on this operational status information, determining the target operation content of the target processing system can include: based on the difference between the current operational status information and the historical operational status information, it can be determined whether the target processing system is operating normally, thereby determining the target operation content of the target processing system. In one possible design, the operational status information can be stored in tables or files to facilitate comparison between the current and historical operational status information, thereby determining the target operation content.

[0214] Optionally, a correlation between operational status information and the operational content of the target processing system can be established in advance. Determining the target operational content of the target processing system based on the operational status information may include: querying the correlation between operational status information and operational content to determine the target operational content corresponding to the current operational status information.

[0215] In some embodiments, determining the target operation content of the target operation information based on the running status information may include:

[0216] Based on the operational status information, determine whether there are operational faults in the target processing system;

[0217] If so, the target operation content is determined to be a fault handling instruction for the target processing system; executing target operations that match the target operation content on the target processing system includes: responding to the fault handling instruction and handling operational faults in the target processing system.

[0218] If not, return to the step of obtaining the updated model parameters for each of the multiple online decision models and continue execution.

[0219] In some embodiments, determining the target operation content of the target processing system based on runtime status information may include:

[0220] Based on the operational status information, the contribution of multiple online decision-making models to the target processing system is scored, and the system contribution scores corresponding to each online decision-making model are obtained.

[0221] Based on the system contribution scores corresponding to multiple online decision-making models, contribution prompts for multiple online decision-making models are generated.

[0222] Performing target operations on the target processing system that match the target operation content includes:

[0223] Output contribution prompts to the system management users of the target processing system.

[0224] As yet another embodiment, the method may further include:

[0225] The system management user of the target processing system outputs updated model parameters for each of the multiple online decision models.

[0226] The system detects parameter adjustment operations triggered by the system administrator for any online decision model and obtains parameter adjustment information input by the system administrator for the online decision model.

[0227] Based on the parameter adjustment information, the model parameters of the online decision-making model are adjusted, and the process returns to the processing objective of the target-based processing system to obtain the decision results corresponding to multiple online decision-making models.

[0228] In practical applications, target processing systems can be mainly used to recommend decision-making objects to target users.

[0229] In one possible design, the decision results generated by any online decision model in the target processing system include: recommendation data determined by the online decision model for multiple candidate objects of the target processing system.

[0230] For ease of use, the MAB (Multi-Armed Bandit) algorithm can be used as an example, and the online decision-making model can be an MAB model. Multiple candidate objects in the target processing system can be considered as the robotic arms (Arms) of the MAB model. The MAB model can predict the winning probabilities of these multiple robotic arms. In this case, the recommendation data determined by the online decision-making model for the multiple candidate objects in the target processing system can be considered the winning probabilities determined by the MAB model for these multiple candidate objects. The method by which the MAB model predicts the winning probabilities of the multiple robotic arms can refer to the probability prediction methods of existing MAB algorithms, and will not be elaborated upon here.

[0231] Based on the decision weights corresponding to multiple online decision-making models, the decision results corresponding to the multiple online decision-making models are weighted to obtain the target decision result, including:

[0232] Determine the recommendation data corresponding to each candidate object in multiple online decision-making models, so as to obtain the recommendation data corresponding to each candidate object in multiple online decision-making models;

[0233] Based on the decision weights corresponding to multiple online decision models, the recommendation data of any candidate object in the multiple online decision models are weighted and calculated to obtain the target recommendation data of the candidate object, thus obtaining the target recommendation data corresponding to multiple candidate objects.

[0234] The target decision result is composed of the target recommendation data corresponding to multiple candidate objects.

[0235] In some embodiments, determining the target recommendation information based on the target decision result may include:

[0236] Based on the target recommendation data corresponding to multiple candidate objects, select at least one target object from the multiple candidate objects;

[0237] Identify target recommendation information consisting of at least one target object.

[0238] refer to Figure 7 This is a flowchart of another embodiment of an information processing method provided in this application. The method may include:

[0239] 701: Detects system access requests initiated by the target user for the online transaction system.

[0240] 702: In response to a system access request, determine the decision results generated by multiple online decision models in the online transaction system.

[0241] 703: Based on the decision results of multiple online decision-making models, determine the target recommendation information corresponding to the recommended transaction object.

[0242] Optionally, an online trading system can be an information system that displays products via web pages and initiates product transactions through these pages. The online trading system can provide product trading services through websites, applications, etc. It can also recommend tradable items to target users. This recommendation information may include details about the recommended items, such as links, names, titles, brief descriptions, images, video introductions, and text descriptions.

[0243] 704: Target user outputs target recommendation information for recommended trading objects.

[0244] In this embodiment, the target processing system may include a network transaction system. When detecting a system access request initiated by a target user for the network transaction system, the system can respond to the access request by determining the decision results generated by multiple online decision models within the network transaction system. Then, using the decision results corresponding to each of the multiple online decision models, target recommendation information corresponding to the recommended transaction object is determined. The recommended transaction object in the target recommendation information can achieve effective object recommendation. The process of recommending transaction objects comprehensively considers the decision results generated by multiple online decision models, achieving integrated recommendation across multiple models, increasing the effectiveness and reliability of the target recommendation information, and improving recommendation efficiency.

[0245] As an example, determining the target recommendation information for recommended transaction objects based on the decision results corresponding to the multiple online decision models may include:

[0246] The decision results corresponding to the multiple online decision-making models are fused together to obtain the target decision result.

[0247] Based on the target decision results, the target recommendation information for the recommended transaction object is determined.

[0248] As yet another example, multiple online decision-making models are determined in the following manner:

[0249] Select at least one objective decision algorithm from multiple online decision algorithms;

[0250] Based on the hyperparameter setting rules corresponding to any target decision algorithm, at least one hyperparameter corresponding to the target decision algorithm is determined; the online decision models corresponding to the at least one hyperparameter are obtained respectively, so as to obtain at least one online decision algorithm corresponding to the target decision algorithm;

[0251] The plurality of online decision models are formed by at least one online decision model for each of the at least one target decision algorithm.

[0252] Figure 7 Some steps in the illustrated embodiment are similar to Figure 1 Some steps are the same in the illustrated embodiments, and for the sake of brevity, they will not be repeated here.

[0253] For ease of understanding, the technical solution of this application embodiment will be described in detail using the information processing method provided in the embodiment of this application configured on a cloud server as an example. (Reference) Figure 8 The technical solution of this application will be described in detail, taking the example of how the target user U can interact with the cloud server M2 through computer M1 to achieve information processing.

[0254] Computer M1 can detect system access requests initiated by target user 801 and send the system access requests to cloud server M2 802. Cloud server M2 is configured with a target processing system and the information processing method provided in this application embodiment.

[0255] After detecting the system access request initiated by the target user for the target processing system, the cloud server M2 can respond to the system access request and determine the decision results generated by more than 803 online decision models in the target processing system.

[0256] Taking the LinTS model, the first LinUCB model, the second LinUCB model, the third LinUCB model, and the fourth LinUCB model in the previous example as examples, we can illustrate this with several online decision-making models. Since these models are all online decision-making models, the decision-making objects of these models can often be referred to as robotic arms. A robotic arm can correspond to the decision-making objects of multiple online decision-making models, i.e., multiple candidate objects. The decision results generated by any online decision-making model in the target processing system can include recommendation data corresponding to multiple robotic arms.

[0257] After receiving decision results from multiple decision systems within the target processing system, cloud server M2 can determine target recommendation information (804) based on the decision results corresponding to multiple online decision models. This target recommendation information is then output to the target user. When outputting target recommendation information, cloud server M2 can send the target recommendation information (805) to computer M1.

[0258] After receiving the target recommendation information, computer M1 can display the target recommendation information (806) to the user, providing the latest target recommendation information to the target user in a timely manner, thus achieving effective recommendations and improving recommendation efficiency. When computer M1 outputs target recommendation information to the target user, the specific output method can include various forms, such as data, pages, information, or messages. This application embodiment does not impose excessive limitations on the specific output method of the target recommendation information.

[0259] like Figure 9 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include the following steps:

[0260] 901: Detects system access requests initiated by target users for the content recommendation system.

[0261] 902: In response to a system access request, determine the decision results generated by multiple online decision models in the content recommendation system.

[0262] 903: Determine the target recommended content based on the decision results corresponding to multiple online decision-making models.

[0263] 904: Outputs target recommended content to the target users.

[0264] A content recommendation system is an information system that provides users with content recommendations and facilitates content viewing via the internet. Specific applications of content recommendation can include: online course recommendations, audio / video cover art recommendations, video title recommendations, and online drug recommendations corresponding to disease diagnoses in the medical field.

[0265] Figure 9 Some steps in the illustrated embodiment are similar to Figure 1 Some steps are the same in the illustrated embodiments, and for the sake of brevity, they will not be repeated here.

[0266] In this embodiment, the target processing system may include a content recommendation system. When a system access request initiated by a target user to the content recommendation system is detected, the system can respond to the access request by determining the decision results generated by multiple online decision models in the content recommendation system, and then using the decision results corresponding to the multiple online decision models, determine the target recommendation information corresponding to the recommended transaction object. The recommended content in the target recommendation information can achieve effective content recommendation. The process of obtaining recommended content comprehensively considers the decision results generated by multiple online decision models, achieving integrated recommendation by multiple models, increasing the effectiveness and reliability of the target recommended content, and improving recommendation efficiency.

[0267] As an example, determining the target recommended content for the recommended transaction object based on the decision results corresponding to the multiple online decision models may include:

[0268] The decision results corresponding to the multiple online decision-making models are fused together to obtain the target decision result.

[0269] Based on the target decision results, the target recommended content for the recommended transaction object is determined.

[0270] As yet another example, multiple online decision-making models are determined in the following manner:

[0271] Select at least one objective decision algorithm from multiple online decision algorithms;

[0272] Based on the hyperparameter setting rules corresponding to any target decision algorithm, at least one hyperparameter corresponding to the target decision algorithm is determined; the online decision models corresponding to the at least one hyperparameter are obtained respectively, so as to obtain at least one online decision algorithm corresponding to the target decision algorithm;

[0273] The plurality of online decision models are formed by at least one online decision model for each of the at least one target decision algorithm.

[0274] like Figure 10 The diagram shown is a flowchart of another embodiment of an information processing method provided in this application. The method may include:

[0275] 1001: In response to a call request to the adversarial decision-making interface, determine the information processing resources corresponding to the adversarial decision-making interface.

[0276] The following steps are performed using the information processing resources corresponding to the adversarial decision-making interface:

[0277] 1002: Utilize the adversarial decision-making interface to detect system access requests initiated by the target user against the target processing system.

[0278] 1003: In response to a system access request, determine the decision results generated by multiple online decision models in the target processing system.

[0279] 1004: Determine target recommendation information based on the decision results corresponding to multiple online decision-making models.

[0280] 1005: Utilize the adversarial decision-making interface to output target recommendation information to the target user.

[0281] In one possible design, the adversarial decision-making interface can be defined as a processing protocol using the technical solutions provided in the embodiments of this application, thereby obtaining an application programming interface (API) that provides software services to the outside world. The adversarial decision-making interface includes interfaces of the type such as SDK (Software Development Kit) and API (Application Programming Interface). During the application of the technical solutions in the embodiments of this application, the interface can be obtained by the user, and the user's system access request can be received through the interface. After obtaining the target recommendation information, the target recommendation information can be output to the user through the interface. Taking a network transaction system as an example, the object information of the product object can be fed back to the user.

[0282] Optionally, the technical solution provided in this application embodiment can be configured in a cloud server. By encapsulating the information processing method, an adversarial decision-making interface that provides the information processing method to the outside world can be called by user devices to provide information processing services to users.

[0283] The specific steps executed by the processing resources corresponding to the adversarial decision-making interface in this application embodiment are... Figure 1 The information processing methods shown perform the same processing steps, and the specific implementation methods and technical effects of each technical feature have been described in [the document / document / etc.]. Figure 1 The embodiments shown are described in detail and will not be repeated here.

[0284] The information processing solutions in this application embodiment are provided to target users in the form of services. The specific service forms include web page services, software program services, or program modules existing in the form of SDKs or APIs. Target users obtain the information processing services provided by this application embodiment by browsing information processing interfaces, service software, or software module interfaces.

[0285] The technical solutions of this application can be applied to various fields such as artificial intelligence interaction, data retrieval, content recommendation, click-through rate prediction, smart factories, and industrial control. They are particularly applicable in the field of content recommendation, such as content recommendation in e-commerce, live video streaming, social networking, and online education, as well as in the field of resource allocation, such as financial product allocation, power resources, water resources, and supply chain allocation.

[0286] To facilitate understanding, the embodiments of this application will be described in detail using the following practical examples of problems in various fields.

[0287] (1) E-commerce scenario

[0288] In e-commerce, feature search, product recommendation in live streaming scenarios, content recommendation, and calculating ad click-through rates are the most common application scenarios. This embodiment takes the content recommendation scenario as an example and provides a deployment example. The general recommendation process in recommendation scenarios can be as follows: First, the recommendation objects for the selected scenario are initially confirmed. During the user's browsing of web pages, the system can determine the target recommendation information based on the decision results generated by multiple online decision models in the target processing system, according to the server providing information processing services. The target recommendation information can then be fed back to the target user.

[0289] (2) Social Scenarios

[0290] In the social networking domain, content recommendation for social network users and material recommendation for students are also quite common. Social network recommendations typically involve users browsing social applications, and the application's interface displaying social content that interests the user. Generally, social network recommendations are based on user information such as browsing history, followed areas, and purchasing behavior. When a target user browses a social page provided by a social network, the user's client can generate a system access request for the target user and send the system access request to a server configured with the technical solution provided in this application.

[0291] When a server receives a system access request from a user, it typically responds by determining the decision results generated by multiple online decision models within the social network. Based on these results, it then determines the target recommendation information. This target recommendation information is obtained through a comprehensive decision-making process involving multiple online decision models on content within the social network, resulting in higher accuracy and the identification of content that is of interest to the social user, thus providing accurate content tailored to that user.

[0292] (3) Material Recommendation Scenarios

[0293] When publishing videos on video websites, it's necessary to set information such as poster images and video screenshots for the target users. Users can trigger a material retrieval request for a specific video; this request is a system access request. During the recommendation process, since different users have different areas of interest, showing users cover images that are more interesting to them can effectively increase video click-through rates and viewing time. The appropriate type is selected from multiple candidate types for a given video to serve as the cover image type for video publishing. In general recommendation scenarios, the technical solution of this application can respond to user-initiated system access requests for cover image types, utilizing the decision results generated by multiple online decision models in the target processing system, and thus determining target recommendation information based on the decision results corresponding to the multiple online decision models. These multiple decision results can be decisions made on multiple candidate types of cover images. The target recommendation information can be the target type obtained from the decisions. The target type can include one or more. After obtaining the target type, it can be fed back to the target user, allowing them to create the video cover image according to the decided target type and complete the video publishing.

[0294] (4) Content delivery scenarios

[0295] In the field of content delivery, promotional content for products can often be delivered in various scenarios, with different scenarios yielding different results. Furthermore, promotional content for products can typically include multiple types, and the effectiveness of different content delivery methods also varies. The technical solution provided in this application can make decisions and recommendations regarding delivery scenarios and / or content.

[0296] To achieve better campaign performance, decisions are typically made regarding multiple candidate scenarios for advertising content. Before content delivery, using the technical solution provided in this application, the content delivery provider can initiate a scenario decision request to a computing device. In response to this request, decision results generated by multiple online decision models within a content click system can be utilized. The content click system simulates user actions such as clicking, purchasing, or browsing the advertised content. The online decision models learn and determine the delivery results for different scenarios within the content click system. Based on the decision results from the multiple online decision models, target recommendation information is determined. In one possible design, any decision result can represent the probability of winning a decision among multiple candidate scenarios. The target recommendation information represents the candidate scenario with the highest winning probability, which is then output to the user as the target recommendation information, allowing the user to deliver the advertising content according to this scenario.

[0297] Furthermore, for different promotional content, target users can typically initiate content decision requests to the content click system. In response to these requests, multiple online decision models can be used to generate decision results within the content click system. The content click system can simulate user actions such as clicking, purchasing, or browsing promotional content; a user clicking on a piece of promotional content is considered a win. The online decision models can evaluate the promotional results of multiple candidate promotional content within the content click system. Based on the decision results from these multiple online decision models, target recommendation information is determined. In one possible design, any decision result can represent the probability of winning for each of the multiple candidate promotional content. The target recommendation information is the candidate promotional content with the highest winning probability, which is then output to the user as the target recommendation information, allowing the user to choose the promotional content recommended by the target recommendation information.

[0298] By making pre-decision decisions for different scenarios or content, effective ad delivery can be achieved, improving campaign performance. The decision-making process employs multiple online decision-making models to synthesize their results, enabling multi-faceted decision-making and enhancing accuracy.

[0299] like Figure 11 The diagram shown is a structural schematic of one embodiment of an information processing device provided in this application. The device may include the following modules:

[0300] Request detection module 1101: Used to detect system access requests initiated by the target user for the target processing system.

[0301] Request response module 1102: Used to respond to system access requests and determine the decision results generated by multiple online decision models in the target processing system.

[0302] Information determination module 1103: used to determine target recommendation information based on the decision results corresponding to multiple online decision models.

[0303] Information output module 1104: Used to output target recommendation information to target users.

[0304] As one embodiment, the information determination module may include:

[0305] The result fusion unit is used to fuse the decision results corresponding to multiple online decision models to obtain the target decision result.

[0306] The recommendation determination unit is used to determine target recommendation information based on the target decision results.

[0307] In some embodiments, the result fusion unit may include:

[0308] The weight determination subunit is used to determine the decision weights corresponding to multiple online decision-making models.

[0309] The result fusion subunit is used to weight the decision results of multiple online decision models according to their respective decision weights to obtain the target decision result.

[0310] In some embodiments, the device may further include:

[0311] The operation acquisition module is used to acquire the feedback operations performed by the target user in response to the target recommendation information;

[0312] The feedback update module is used to update the model parameters of multiple online decision-making models based on the feedback data corresponding to the feedback operation, so as to obtain multiple online decision-making models with updated parameters.

[0313] The decision update module is used to obtain the updated decision results corresponding to multiple online decision models and then jump to the information determination module to continue execution.

[0314] In some embodiments, the result fusion unit may further include:

[0315] The weight update subunit is used to update the decision weights of multiple online decision models based on the feedback data corresponding to the feedback operation, so as to return to the result fusion subunit for continued execution.

[0316] In some embodiments, the device may further include:

[0317] The weight output unit is used to output the updated decision weights for each of the multiple online decision models to the system management user.

[0318] In one possible design, the device may also include:

[0319] The information acquisition module is used to detect the weight adjustment operations triggered by the system management user of the target processing system on the decision weights corresponding to multiple online decision models, and to acquire the weight adjustment information input by the system management user.

[0320] The weight adjustment module is used to update the decision weights corresponding to multiple online decision models according to the weight adjustment information, obtain the updated decision weights corresponding to the multiple online decision models, and return to continue execution.

[0321] As yet another embodiment, the device may further include:

[0322] The model determination module is used to obtain the model parameters of each of the updated online decision models.

[0323] The convergence acquisition module is used to determine the model convergence information corresponding to the updated model parameters of any online decision model, so as to obtain the model convergence information corresponding to multiple online decision models respectively.

[0324] The state determination module is used to generate the operating state information of the target processing system based on the model convergence information corresponding to multiple online decision models.

[0325] The operation determination module is used to determine the target operation content of the target processing system based on the running status information.

[0326] The system operation module is used to execute target operations on the target processing system that match the target operation content.

[0327] In one possible design, the operation determination module may include:

[0328] The fault diagnosis unit is used to determine whether there is an operational fault in the target processing system based on the operating status information.

[0329] The first processing unit is used to determine, if so, that the target operation content is a fault handling instruction for the target processing system. The system operation module may include: an instruction control unit, used to handle operational faults in the target processing system in response to the fault handling instruction.

[0330] The second processing unit is used to return to the step of obtaining the updated model parameters of the multiple online decision models if no, and continue execution.

[0331] In another possible design, the operation determination module may include:

[0332] The score determination unit is used to score the contribution of multiple online decision-making models to the target processing system based on the running status information, and to obtain the system contribution score corresponding to each of the multiple online decision-making models.

[0333] The contribution prompting unit is used to generate contribution prompting information for multiple online decision-making models based on the system contribution scores corresponding to each model.

[0334] As another embodiment, the decision results generated by any online decision model in the target processing system may include: recommendation data determined by the online decision model for multiple candidate objects of the target processing system.

[0335] The result fusion subunit may include:

[0336] The data determination module is used to determine the recommendation data corresponding to each candidate object in multiple online decision models, so as to obtain the recommendation data corresponding to each candidate object in multiple online decision models.

[0337] The data weighting module is used to calculate the target recommendation data of any candidate object based on the decision weights of the multiple online decision models, thereby obtaining the target recommendation data of the candidate object.

[0338] The result determination module is used to determine the target decision result composed of the target recommendation data corresponding to multiple candidate objects.

[0339] In some embodiments, the recommendation determination unit may include:

[0340] The object selection subunit is used to select at least one target object from multiple candidate objects based on the target recommendation data corresponding to each candidate object.

[0341] The information determination subunit is used to determine target recommendation information consisting of at least one target object.

[0342] As yet another embodiment, the device may further include:

[0343] The algorithm selection module is used to select at least one target decision algorithm from multiple online decision algorithms.

[0344] The parameter setting module is used to obtain at least one online decision model corresponding to the target decision algorithm, so as to obtain multiple online decision models composed of at least one online decision model of each of the target decision algorithms.

[0345] In one possible design, the algorithm selection module may include:

[0346] System feature unit, used to determine the system feature information of the target processing system.

[0347] The first determining unit is used to select at least one target decision algorithm from multiple online decision algorithms based on system characteristic information.

[0348] In some embodiments, the first determining unit may include:

[0349] The first matching subunit is used to determine the first matching degree between multiple online decision-making algorithms and system feature information.

[0350] The first decision subunit is used to select at least one target decision algorithm from multiple online decision algorithms according to the first matching degree corresponding to each of the multiple online decision algorithms.

[0351] In some embodiments, the algorithm selection module may include:

[0352] User feature unit, used to determine the user feature information of the target user.

[0353] The second determining unit is used to select at least one target decision algorithm from multiple online decision algorithms based on user characteristic information.

[0354] In one possible design, the second determining unit may include:

[0355] The second matching subunit is used to determine the second matching degree between multiple online decision-making algorithms and user feature information respectively;

[0356] The second decision subunit is used to select at least one target decision algorithm from multiple online decision algorithms according to the second matching degree corresponding to each of the multiple online decision algorithms.

[0357] As one possible implementation, the algorithm selection module may include:

[0358] The algorithm output unit is used to output the multiple online decision-making algorithms to the system management user.

[0359] The user selection unit is used to obtain the target decision algorithm selected by the system management user from the plurality of online decision algorithms.

[0360] In some embodiments, at least one online decision model corresponding to any objective decision algorithm is determined in the following way:

[0361] Determine the hyperparameter setting rules for any objective decision-making algorithm;

[0362] Based on the hyperparameter setting rules, at least one hyperparameter of the target decision algorithm is determined;

[0363] Obtain the online decision models corresponding to the at least one hyperparameter, so as to obtain at least one online decision model corresponding to the target decision algorithm.

[0364] Figure 11 The information processing device can perform Figure 1 The implementation principle and technical effects of the information processing method in the illustrated embodiment will not be repeated here. The specific manner in which each module, unit, and subunit in the above embodiments perform each step has been described in detail in the embodiments related to the method, and will not be elaborated here.

[0365] In practical applications, Figure 11 The information processing device shown can be configured as a computing device. For example... Figure 12The diagram shown is a structural schematic of one embodiment of a computing device provided in this application. The device may include: a storage component 1201 and a processing component 1202; the storage component 1201 is used to store one or more computer instructions; one or more computer instructions are invoked by the processing component 1202 to execute. Figures 1-10 Any of the information processing methods shown.

[0366] The processing component 1202 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0367] Storage component 1201 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0368] Of course, computing devices may also include other components, such as input / output interfaces and communication components. Input / output interfaces provide an interface between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between the computing device and other devices.

[0369] Furthermore, embodiments of this application also provide a computer-readable storage medium that can store one or more computer instructions, which, when executed, are used to implement any of the information processing methods in the embodiments of this application.

[0370] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0371] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of a necessary general-purpose hardware platform, or by a combination of hardware and software. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a computer product. This application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0372] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An information processing method, characterized in that, include: Detect system access requests initiated by the target user against the target processing system; In response to the system access request, the decision results generated by multiple online decision models in the target processing system are determined respectively; wherein, the multiple online decision models are used to generate decision results respectively; the multiple online decision models include online decision models with the same algorithm type but different model hyperparameters; Based on the decision results corresponding to the multiple online decision models, target recommendation information is determined. Output the target recommendation information to the target user; The determination of target recommendation information based on the decision results corresponding to the multiple online decision models includes: The decision results corresponding to the multiple online decision-making models are fused together to obtain the target decision result. Based on the target decision result, the target recommendation information is determined.

2. The method according to claim 1, characterized in that, The step of fusing the decision results corresponding to the multiple online decision models to obtain the target decision result includes: Determine the decision weights corresponding to the multiple online decision models; Based on the decision weights corresponding to the multiple online decision models, the decision results corresponding to the multiple online decision models are weighted to obtain the target decision result.

3. The method according to claim 2, characterized in that, After outputting the target recommendation information to the target user, the method further includes: Obtain the feedback operation performed by the target user in response to the target recommendation information; Based on the feedback data corresponding to the feedback operation, update the model parameters of each of the multiple online decision-making models to obtain multiple online decision-making models with updated parameters. The updated decision results corresponding to the multiple online decision models are obtained and returned to the step of determining the target recommendation information based on the decision results corresponding to the multiple online decision models.

4. The method according to claim 3, characterized in that, Also includes: Based on the feedback data corresponding to the feedback operation, the decision weights corresponding to the multiple online decision models are updated respectively, so as to return to the step of weighting the decision results corresponding to the multiple online decision models according to their respective decision weights to obtain the target decision result.

5. The method according to claim 3, characterized in that, After updating the model parameters of each of the multiple online decision-making models based on the feedback data corresponding to the feedback operation, and obtaining the multiple online decision-making models with updated parameters, the process further includes: Obtain the updated model parameters for each of the multiple online decision-making models; For any online decision-making model, determine the model convergence information corresponding to the updated model parameters of the online decision-making model, so as to obtain the model convergence information corresponding to the multiple online decision-making models respectively; Based on the model convergence information corresponding to the multiple online decision models, the operating status information of the target processing system is generated. Based on the operational status information, the target operation content of the target processing system is determined; The target processing system performs a target operation that matches the target operation content.

6. The method according to claim 5, characterized in that, The determination of the target operation content of the target processing system based on the operating status information includes: Based on the operational status information, it is determined whether the target processing system has any operational faults; If so, the target operation content is determined to be a fault handling instruction of the target processing system; the execution of a target operation matching the target operation content on the target processing system includes: in response to the fault handling instruction, handling the operational fault in the information target processing system; If not, return to the step of obtaining the updated model parameters of each of the multiple online decision models and continue execution.

7. The method according to claim 5, characterized in that, The determination of the target operation content of the target processing system based on the operating status information includes: Based on the operational status information, the contribution of the multiple online decision-making models to the target processing system is scored to obtain the system contribution scores corresponding to the multiple online decision-making models. Based on the system contribution scores corresponding to the multiple online decision-making models, contribution prompt information for the multiple online decision-making models is generated. The execution of a target operation on the target processing system that matches the target operation content includes: The contribution prompt information is output to the system management user of the target processing system.

8. The method according to claim 5, characterized in that, Also includes: The updated model parameters of each of the multiple online decision models are output to the system management user of the target processing system. The system detects parameter adjustment operations triggered by the system management user for any online decision model and obtains parameter adjustment information input by the system management user for the online decision model. Based on the parameter adjustment information, the model parameters of the online decision-making model are adjusted, and the process returns to the processing target based on the target processing system to obtain the decision results corresponding to the multiple online decision-making models.

9. The method according to claim 1, characterized in that, The multiple online decision-making models were determined in the following way: Select at least one objective decision algorithm from multiple online decision algorithms; Obtain at least one online decision model corresponding to any target decision algorithm, and obtain multiple online decision models composed of at least one online decision model of each of the at least one target decision algorithms.

10. The method according to claim 9, characterized in that, The step of selecting at least one target decision algorithm from multiple online decision algorithms includes: Determine the system characteristic information of the target processing system; Based on the system feature information, at least one target decision algorithm is selected from the plurality of online decision algorithms.

11. The method according to claim 10, characterized in that, The step of selecting the at least one target decision algorithm from the plurality of online decision algorithms based on the system feature information includes: Determine the first matching degree between the plurality of online decision-making algorithms and the system feature information respectively; Based on the first matching degree corresponding to the plurality of online decision algorithms, at least one target decision algorithm is selected from the plurality of online decision algorithms.

12. The method according to claim 11, characterized in that, The step of selecting at least one target decision algorithm from multiple online decision algorithms includes: Determine the user characteristic information of the target user; Based on the user characteristic information, at least one target decision algorithm is selected from the plurality of online decision algorithms.

13. The method according to claim 12, characterized in that, The step of selecting the at least one target decision algorithm from the plurality of online decision algorithms based on the user feature information includes: Determine the second matching degree between the plurality of online decision algorithms and the user feature information; Based on the second matching degree corresponding to each of the multiple online decision algorithms, at least one target decision algorithm is selected from the multiple online decision algorithms.

14. The method according to claim 9, characterized in that, The step of selecting at least one target decision algorithm from multiple online decision algorithms includes: The system management user is provided with the aforementioned multiple online decision-making algorithms; Obtain the at least one target decision algorithm selected by the system management user from the plurality of online decision algorithms.

15. The method according to claim 9, characterized in that, At least one online decision model corresponding to any given objective decision algorithm is determined in the following way: Determine the hyperparameter setting rules for any objective decision-making algorithm; Based on the hyperparameter setting rules, at least one hyperparameter of the target decision algorithm is determined; Obtain the online decision models corresponding to the at least one hyperparameter, so as to obtain at least one online decision model corresponding to the target decision algorithm.

16. An information processing method, characterized in that, include: Detect system access requests initiated by target users against the online transaction system; In response to the system access request, the decision results generated by multiple online decision models in the online transaction system are determined respectively; wherein, the multiple online decision models are used to generate decision results respectively; the multiple online decision models include online decision models with the same algorithm type but different model hyperparameters; Based on the decision results corresponding to the multiple online decision models, the target recommendation information for the recommended transaction object is determined. Output the target recommendation information of the recommended transaction object to the target user; The determination of target recommendation information for recommended transaction objects based on the decision results corresponding to the multiple online decision models includes: The decision results corresponding to the multiple online decision-making models are fused together to obtain the target decision result. Based on the target decision result, the target recommendation information is determined.

17. The method according to claim 16, characterized in that, The multiple online decision-making models were determined in the following way: Select at least one objective decision algorithm from multiple online decision algorithms; Based on the hyperparameter setting rules corresponding to any target decision algorithm, determine at least one hyperparameter corresponding to the target decision algorithm; Obtain the online decision models corresponding to the at least one hyperparameter, so as to obtain at least one online decision algorithm corresponding to the target decision algorithm; The plurality of online decision models are formed by at least one online decision model for each of the at least one target decision algorithm.

18. An information processing method, characterized in that, include: Detect system access requests initiated by target users for the content recommendation system; In response to the system access request, the decision results generated by multiple online decision models in the content recommendation system are determined respectively; wherein, the multiple online decision models are used to generate decision results respectively; the multiple online decision models include online decision models with the same algorithm type but different model hyperparameters; Based on the decision results corresponding to the multiple online decision models, the target recommended content is determined. Output the target recommended content to the target user; The determination of the target recommendation content based on the decision results corresponding to the multiple online decision models includes: The decision results corresponding to the multiple online decision-making models are fused together to obtain the target decision result. Based on the target decision results, the target recommended content is determined.

19. An information processing method, characterized in that, include: In response to a call request to the adversarial decision-making interface, determine the information processing resources corresponding to the adversarial decision-making interface; The following steps are performed using the information processing resources corresponding to the adversarial decision-making interface: The adversarial decision interface is used to detect system access requests initiated by the target user against the target processing system; In response to the system access request, the decision results generated by multiple online decision models in the target processing system are determined respectively; wherein, the multiple online decision models are used to generate decision results respectively; the multiple online decision models include online decision models with the same algorithm type but different model hyperparameters; Based on the decision results corresponding to the multiple online decision models, target recommendation information is determined. The target recommendation information is output to the target user using the adversarial decision interface. The determination of target recommendation information based on the decision results corresponding to the multiple online decision models includes: The decision results corresponding to the multiple online decision-making models are fused together to obtain the target decision result. Based on the target decision result, the target recommendation information is determined.

20. An information processing device, characterized in that, include: The request detection module is used to detect system access requests initiated by target users for the target processing system; The request-response module is used to respond to the system access request and determine the decision results generated by multiple online decision models in the target processing system; wherein, the multiple online decision models are used to generate decision results respectively; the multiple online decision models include online decision models with the same algorithm type but different model hyperparameters; The information determination module is used to determine target recommendation information based on the decision results corresponding to the multiple online decision models. The information output module is used to output the target recommendation information to the target user; The information determination module is specifically used to: perform result fusion processing on the decision results corresponding to the multiple online decision models respectively to obtain the target decision result; and determine the target recommendation information based on the target decision result.

21. A computing device, characterized in that, It includes: a storage component and a processing component; the storage component is used to store one or more computer instructions; the one or more computer instructions are invoked by the processing component to execute the information processing method according to any one of claims 1 to 15.

Images