Method and apparatus for updating recommendation model based on large model, device and medium

Abstract

The present disclosure provides a large model-based recommendation model updating method and device, equipment and medium, relating to the technical field of artificial intelligence, and particularly to the technical field of large models, agents and recommendations. The implementation scheme is: determining a first prompt word based on a to-be-optimized index of a recommendation model, current parameter values of a plurality of parameter items of the recommendation model, and a current index value of the to-be-optimized index; calling a large model using the first prompt word, so that the large model generates a plurality of first parameter groups; determining a first index value corresponding to the to-be-optimized index of each first parameter group based on interaction information of a user on a first output result of a recommendation model corresponding to the first parameter group; determining a second prompt word based on the first prompt word, the plurality of first parameter groups, and the first index values corresponding to the plurality of first parameter groups respectively; calling the large model using the second prompt word, so that the large model generates at least one second parameter group; and updating the recommendation model based on the at least one second parameter group.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more particularly to the fields of large models, intelligent agents, and recommendation technology, specifically to a method, apparatus, electronic device, computer-readable storage medium, and computer program product for updating a recommendation model based on a large model. Background Technology

[0002] Artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies mainly include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0003] Recommendation systems typically use recommendation models to output information streams consisting of multiple content blocks when distributing personalized content to users. In related technologies, recommendation models are trained and optimized by human engineers based on experience, resulting in high human costs. Furthermore, the iterative update efficiency of recommendation models is low, thus affecting recommendation efficiency and accuracy.

[0004] The methods described in this section are not necessarily methods that had been previously conceived or adopted. Unless otherwise specified, no method described in this section should be assumed to be prior art simply because it is included in this section. Similarly, unless otherwise specified, the issues mentioned in this section should not be considered to be accepted in any prior art. Summary of the Invention

[0005] This disclosure provides a method, apparatus, electronic device, computer-readable storage medium, and computer program product for updating a recommendation model based on a large model.

[0006] According to one aspect of this disclosure, a method for updating a recommendation model based on a large model is provided, wherein the recommendation model includes multiple parameter items to be updated, and the method includes: determining a first prompt word based on an indicator to be optimized for the recommendation model, current parameter values ​​of the multiple parameter items, and current indicator values ​​of the indicator to be optimized, wherein the indicator to be optimized includes an indicator characterizing user interaction information with the output results of the recommendation model; invoking the large model using the first prompt word, such that the large model generates multiple first parameter groups, wherein each of the multiple first parameter groups includes multiple candidate parameter values ​​corresponding to the multiple parameter items respectively; For each of the plurality of first parameter groups, based on the user's interaction information with the corresponding first output result of each of the plurality of first parameter groups, a first indicator value corresponding to the indicator to be optimized for that first parameter group is determined, wherein the first output result is obtained using the recommendation model configured based on that first parameter group; a second prompt word is determined based on the first prompt word, the plurality of first parameter groups, and the corresponding first indicator value of each of the plurality of first parameter groups; the large model is invoked using the second prompt word so that the large model generates at least one second parameter group; and the recommendation model is updated based on the at least one second parameter group.

[0007] According to another aspect of this disclosure, an update apparatus for a recommendation model based on a large model is provided, wherein the recommendation model includes multiple parameter items to be updated, and the apparatus includes: a first determining unit configured to determine a first prompt word based on an indicator to be optimized for the recommendation model, current parameter values ​​of the multiple parameter items, and current indicator values ​​of the indicator to be optimized, wherein the indicator to be optimized includes an indicator characterizing user interaction information with the output results of the recommendation model; a first invoking unit configured to invoke the large model using the first prompt word, such that the large model generates multiple first parameter groups, wherein each of the multiple first parameter groups includes multiple candidate parameter values ​​corresponding to the multiple parameter items respectively; and a second determining unit configured to... For each of the plurality of first parameter groups, based on the user's interaction information with the corresponding first output result of each of the plurality of first parameter groups, a first indicator value corresponding to the indicator to be optimized for that first parameter group is determined, wherein the first output result is obtained using the recommendation model configured based on that first parameter group; a third determining unit is configured to determine a second prompt word based on the first prompt word, the plurality of first parameter groups, and the corresponding first indicator value of each of the plurality of first parameter groups; a second calling unit is configured to call the large model using the second prompt word, so that the large model generates at least one second parameter group; and an updating unit is configured to update the recommendation model based on the at least one second parameter group.

[0008] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method for updating the recommendation model based on a large model.

[0009] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the above-described method for updating the recommendation model based on a large model.

[0010] According to another aspect of this disclosure, a computer program product is provided, including a computer program, wherein the computer program, when executed by a processor, is capable of implementing the above-described method for updating the recommendation model based on a large model.

[0011] According to one or more embodiments of this disclosure, the update efficiency of the recommendation model can be improved, and the recommendation accuracy can be enhanced.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings exemplify embodiments and form part of the specification, serving together with the textual description to explain exemplary implementations of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. Throughout the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0014] Figure 1 A schematic diagram of an exemplary system in which various methods described herein may be implemented, according to exemplary embodiments of the present disclosure; Figure 2 A flowchart is shown of an update method for a recommendation model based on a large model according to an exemplary embodiment of the present disclosure; Figure 3 A schematic diagram illustrating the update process of a large-model-based recommendation model according to an exemplary embodiment of the present disclosure is shown. Figure 4 A structural block diagram of an apparatus for updating a recommendation model based on a large model according to an exemplary embodiment of the present disclosure is shown. Figure 5 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0015] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0016] In this disclosure, unless otherwise stated, the use of terms such as "first," "second," etc., to describe various elements is not intended to limit the positional, temporal, or importance relationships of these elements; such terms are merely used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of that element, while in other cases, based on the context, they may refer to different instances.

[0017] The terminology used in the description of the various examples described in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context explicitly indicates otherwise, an element may be one or more unless the number of elements is specifically limited. Furthermore, the term "and / or" as used in this disclosure covers any one of the listed items and all possible combinations thereof.

[0018] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0019] Figure 1 A schematic diagram of an exemplary system 100 in which the various methods and apparatus described herein can be implemented according to embodiments of this disclosure is shown. Reference Figure 1 The system 100 includes one or more client devices 101, 102, 103, 104, 105 and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105 and 106 can be configured to execute one or more applications.

[0020] In embodiments of this disclosure, server 120 may run one or more services or software applications that enable the execution of update methods for recommendation models based on large models.

[0021] In some embodiments, server 120 may also provide other services or software applications, which may include non-virtual and virtual environments. In some embodiments, these services may be provided as web-based services or cloud services, such as to users of client devices 101, 102, 103, 104, 105 and / or 106 under a Software as a Service (SaaS) model.

[0022] exist Figure 1 In the configuration shown, server 120 may include one or more components that implement the functions performed by server 120. These components may include software components, hardware components, or combinations thereof that can be executed by one or more processors. Users operating client devices 101, 102, 103, 104, 105, and / or 106 can sequentially interact with server 120 using one or more client applications to utilize the services provided by these components. It should be understood that various different system configurations are possible and may differ from system 100. Therefore, Figure 1 This is an example of a system used to implement the various methods described herein, and is not intended to be limiting.

[0023] Users can use client devices 101, 102, 103, 104, 105, and / or 106 to send model invocation requests. The client devices can provide an interface that allows users to interact with the client devices. The client devices can also output information to the user through this interface. Although... Figure 1 Only six client devices are described, but those skilled in the art will understand that this disclosure can support any number of client devices.

[0024] Client devices 101, 102, 103, 104, 105, and / or 106 may include various categories of computer devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptops), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors, or other sensing devices. These computer devices can run various categories and versions of software applications and operating systems, such as Microsoft Windows, Apple iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as Google Chrome OS); or include various mobile operating systems, such as Microsoft Windows Mobile OS, iOS, Windows Phone, and Android. Portable handheld devices may include cellular phones, smartphones, tablets, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. Gaming systems may include various handheld gaming devices, internet-enabled gaming devices, etc. Client devices are capable of executing various applications, such as various internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and can use various communication protocols.

[0025] Network 110 can be any type of network well known to those skilled in the art, and can support data communication using any of a variety of available protocols (including, but not limited to, TCP / IP, SNA, IPX, etc.). By way of example only, one or more networks 110 can be a local area network (LAN), an Ethernet-based network, a token ring network, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a blockchain network, a public switched telephone network (PSTN), an infrared network, a wireless network (e.g., Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0026] Server 120 may include one or more general-purpose computers, special-purpose server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running a virtual operating system, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices for servers). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0027] The computing unit in server 120 can run one or more operating systems, including any of the aforementioned operating systems and any commercially available server operating system. Server 120 can also run any of a variety of additional server applications and / or middleware applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0028] In some implementations, server 120 may include one or more applications to analyze and merge data feeds and / or event updates received from users of client devices 101, 102, 103, 104, 105, and 106. Server 120 may also include one or more applications to display data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105, and 106.

[0029] In some implementations, server 120 can be a server for a distributed system or a server integrated with blockchain. Server 120 can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. A cloud server is a host product in the cloud computing service system, designed to address the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0030] System 100 may also include one or more databases 140. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 140 may be used to store information such as audio files and video files. Databases 140 may reside in various locations. For example, a database used by server 120 may be local to server 120, or it may be located away from server 120 and may communicate with server 120 via a network-based or dedicated connection. Databases 140 may be of different categories. In some embodiments, the database used by server 120 may be, for example, a relational database. One or more of these databases may store, update, and retrieve data from and from the databases in response to commands.

[0031] In some embodiments, one or more of the databases 140 may also be used by an application to store application data. The databases used by the application may be different categories of databases, such as key-value stores, object stores, or regular stores supported by a file system.

[0032] Figure 1The system 100 can be configured and operated in various ways to enable the application of the various methods and apparatus described in this disclosure.

[0033] Recommendation systems typically use recommendation models to output information streams consisting of multiple content blocks when distributing personalized content to users. In related technologies, recommendation models are trained and optimized by human engineers based on experience, resulting in high human costs. Furthermore, the iterative update efficiency of recommendation models is low, thus affecting recommendation efficiency and accuracy.

[0034] Based on this, this disclosure provides a method for automatically iteratively updating the parameters of a recommendation model using a large model. The method involves constructing the model optimization objective, the current values ​​of the model parameters, and the current performance of the model as a first prompt word to convey the parameter optimization task information to the large model. The large model then generates multiple candidate parameter combinations, which are configured into the recommendation model for user interaction verification. Based on user feedback, the evaluation value of each candidate parameter group is determined. The large model is then invoked again to generate optimized parameter groups based on the evaluation information of each candidate parameter group, and these optimized parameter groups are used to update the recommendation model. Thus, automatic updating of recommendation model parameters can be achieved based on a closed loop of "candidate generation - evaluation - feedback iteration," improving the recommendation accuracy and iteration efficiency of the recommendation model.

[0035] Figure 2 A flowchart of an update method 200 for a large-model-based recommendation model according to an exemplary embodiment of the present disclosure is shown. The recommendation model includes multiple parameter items to be updated. Figure 2 As shown, method 200 includes: Step S201: Determine the first prompt word based on the indicators to be optimized for the recommendation model, the current parameter values ​​of multiple parameter items, and the current indicator values ​​of the indicators to be optimized. The indicators to be optimized include indicators that characterize the user's interaction information with the output results of the recommendation model. Step S202: Use the first prompt word to call the large model so that the large model generates multiple first parameter groups, wherein each of the multiple first parameter groups includes multiple candidate parameter values ​​corresponding to multiple parameter items respectively; Step S203: For each of the multiple first parameter groups, based on the user's interaction information with the first output result of each of the multiple first parameter groups, determine the first indicator value of the first parameter group corresponding to the indicator to be optimized, wherein the first output result is obtained using the recommendation model configured based on the first parameter group; Step S204: Determine the second prompt word based on the first prompt word, multiple first parameter groups, and the corresponding first index values ​​of each of the multiple first parameter groups; Step S205: Invoke the large model using the second prompt word, so that the large model generates at least one second parameter set; and Step S206: Update the recommendation model based on at least one second parameter group.

[0036] By applying the method described above 200, a large model can be used to automatically iteratively update the parameters of the recommendation model. The first prompt word is constructed from the model's target metric, the current values ​​of the model's parameters, and the model's current performance (current metric value) to convey the parameter optimization task information to the large model. This allows the large model to generate multiple candidate parameter combinations, which are then configured into the recommendation model for user interaction verification. The metric evaluation value for each candidate parameter group is determined based on user feedback. The large model is then invoked again to generate optimized parameter groups based on the metric evaluation information of each candidate parameter group, and these optimized parameter groups are used to update the recommendation model. Thus, automatic updating of the recommendation model parameters can be achieved based on a closed loop of "candidate generation - metric evaluation - feedback iteration," allowing the parameter update process to fully utilize the semantic understanding and logical reasoning capabilities of the large model, thereby improving the recommendation accuracy and iteration efficiency of the recommendation model.

[0037] In some examples, the multiple parameters to be updated in the recommendation model are either all the parameters of the recommendation model or only some of them, such as some key parameters in the model's output layer. In one example, the recommendation model refers to the reordering model applied by the recommendation system. The reordering model is used to merge the predicted scores of multiple dimensions corresponding to multiple content blocks to be recommended into a single score and sort them, that is, to determine the display order of multiple content blocks to be recommended in the information stream. For each content block, the recommendation system first uses the preceding model to output the predicted information of its corresponding multiple dimensions (such as the probability of a user clicking on the content block, the duration of a user viewing the content block, the probability of a user liking or commenting, the probability of a user being satisfied with the recommendation result, etc.). The reordering model is used to perform a fusion calculation based on the importance coefficients that correspond one-to-one with the multiple sets of predicted information and the multiple sets of predicted information for each content block to obtain the corresponding comprehensive score for each content block to be recommended. Then, the multiple content blocks to be recommended are sorted according to this to generate the final recommended content sequence. In this example, the aforementioned importance coefficients are the multiple parameter items that method 200 needs to update. By updating multiple parameter items, the relative importance of each prediction information in the ranking decision can be adjusted more accurately to affect the ranking results output by the recommendation model, thereby improving the recommendation accuracy.

[0038] Metrics to be optimized are quantitative standards used to measure the quality of the recommendation model's output. These metrics can include one or more sub-metrics defined based on the actual interactive behaviors of users after encountering recommended content. For example, when Method 200 is applied to a video feed recommendation scenario, the metrics to be optimized can be one or more of the following: viewing time, completion rate, satisfaction rate, quick scrolling rate, and interaction frequency. User interaction information regarding the recommendation model's output refers to the specific operational data of users on the recommended content sequence, including but not limited to clicking on content blocks, viewing time of content blocks, liking, commenting, sharing, or disliking content blocks, and quickly scrolling through content blocks. By collecting and statistically analyzing various interaction information, the user's preference and satisfaction with the recommendation results can be quantitatively represented, thereby evaluating the accuracy of the recommendation model's output. In some examples, metrics to be optimized can include interaction metrics used to represent users' long-term interests, such as user retention rates based on long-term statistical periods, or long-term user value information obtained by converting or mapping the aforementioned short-term metrics, so that the updated recommendation model's output can better align with users' long-term interests and improve long-term recommendation performance.

[0039] In some examples, the first prompt is instruction text used to convey the recommendation model parameter optimization task to the large model. According to the method provided in this disclosure, the first prompt includes at least the following: a description of the metric to be optimized, used to inform the large model what the optimization goal is, such as "the optimization goal is to increase the total browsing time of users"; information on the current parameter values, used to inform the large model of the specific values ​​of each parameter used in the current recommendation model; and information on the current metric values, used to inform the large model of the actual performance of each metric to be optimized under the current parameter configuration, such as "the average total browsing time per user is X seconds under the current configuration". In some examples, the first prompt further includes task description information to instruct the large model to generate candidate parameter sets, such as "Please generate multiple candidate parameter sets corresponding to multiple parameter items to be updated based on the above information".

[0040] By inputting the first suggestion word into a large model, the model can be used as a parameter optimization engine to generate multiple alternative sets of first parameters. In one example, the large model can generate multiple diverse sets of candidate parameter values ​​based on several different search directions. For instance, the large model might generate multiple differentiated candidate parameter sets based on different strategies such as "significantly adjusting the value of one parameter while keeping the values ​​of other parameters unchanged," "simultaneously fine-tuning the values ​​of multiple parameters to maintain balance," and "adjusting the value of a parameter in different directions." This achieves broad coverage of the parameter space, reduces the probability of getting trapped in local optima, and improves the accuracy of recommendation model updates.

[0041] In some examples, after the large model generates multiple first parameter groups, step S203 can configure the candidate parameter values ​​of each first parameter group into the recommendation model, so that the recommendation model generates the first output result based on the parameters of that group. Then, small-scale tests can be performed on each first output result to collect and statistically analyze user interaction information (such as click data, viewing time data, interaction data, etc.) for each first output result, thereby obtaining the corresponding first indicator value for each first parameter group (such as calculating the specific value of user browsing time or click rate for each first output result) to more accurately evaluate the model recommendation effect of different candidate parameter groups.

[0042] According to step S204 above, a second prompt is constructed by further integrating the evaluation information of each first parameter group and its corresponding indicators based on the first prompt. In some examples, the system can instruct the large model to refer to the task framework defined by the first prompt in natural language in the second prompt. For example, adding a statement such as "Please review the parameter optimization task you received previously and continue parameter optimization" at the beginning of the second prompt can guide the large model to make further optimization decisions by associating with the context. In other examples, the system can explicitly include the corresponding content of the first prompt (e.g., the indicator to be optimized, the current parameter values ​​of multiple parameter items, and the current indicator value of the indicator to be optimized) in the second prompt, and append the first parameter group and its corresponding first indicator value to further instruct the large model to analyze the parameter adjustment strategy based on the indicator evaluation results, so as to generate at least one optimized second parameter group, so that the model performance (indicator value of the indicator to be optimized) of at least one second parameter group is better than the model performance of multiple first parameter groups.

[0043] According to step S205 above, the large model further considers and reasones based on the second prompt word to obtain at least one second parameter group. For example, the large model can compare and analyze the corresponding first indicator values ​​of multiple first parameter groups to identify the first parameter group that improves the first indicator value relative to the current indicator value. Based on this, it can analyze successful parameter adjustment strategies. It can also simultaneously identify parameter groups that cause the indicator value to decrease to analyze failed parameter adjustment strategies. Then, based on the comprehensive analysis results, it generates at least one second parameter group. The second parameter group is used to represent the adjustment results of successful parameter adjustment strategies, thereby obtaining a parameter group used to improve the recommendation effect of the recommendation model.

[0044] Once at least one set of second parameters is obtained, multiple parameters of the recommendation model can be updated. For example, the values ​​of currently used parameters in the recommendation model can be directly replaced based on the second parameter set to obtain an optimized recommendation model. In some examples, when a large model generates multiple sets of second parameters, further small-scale validation can be performed based on these sets to select the second parameter set with the best performance metrics as the final parameter set for updating the recommendation model, thereby improving the output accuracy of the recommendation model.

[0045] According to some embodiments, step S202, which calls the large model using the first prompt word so that the large model generates multiple first parameter groups, includes: calling the large model using the first prompt word so that the large model generates multiple first parameter groups based on multiple first evolutionary strategies, wherein the large model performs the following operations based on the second prompt word: determining at least one second evolutionary strategy from the multiple first evolutionary strategies based on the first comparison result of the corresponding first index value of each of the multiple first parameter groups and the current index value; and generating at least one second parameter group based on at least one second evolutionary strategy.

[0046] By generating multiple candidate sets of first parameters based on various evolutionary strategies, the diversity of these sets can be enhanced, achieving broad coverage of the parameter space, reducing the probability of getting trapped in local optima, and improving the accuracy of recommendation model updates. Furthermore, by analyzing the comparison results between each first indicator value and the current indicator value (i.e., analyzing whether each first indicator value is better than the current first indicator value), the effectiveness of each evolutionary strategy can be easily and accurately indicated, achieving more efficient and accurate parameter optimization and improving the efficiency and accuracy of recommendation model parameter updates.

[0047] In some examples, multiple evolutionary strategies are different search directions or heuristics with different focuses followed by large models when generating candidate parameter sets. In other examples, besides the above-mentioned method of generating candidate parameter sets based on evolutionary strategies, large models can also employ other types of generation methods, such as using other numerical optimization algorithms for parameter space search.

[0048] According to some embodiments, method 200 further includes: determining a second indicator value corresponding to the indicator to be optimized for each of the at least one second parameter group based on user interaction information of the second output results corresponding to the at least one second parameter group; determining a third prompt word based on the indicator to be optimized, the at least one second parameter group, and the second indicator values ​​of the at least one second parameter group; invoking a large model using the third prompt word, such that the large model generates multiple third parameter groups based on multiple third evolutionary strategies associated with the second evolutionary strategy; determining a third indicator value corresponding to the indicator to be optimized for each third parameter group based on user interaction information of the third output results corresponding to each third parameter group; determining at least one fourth parameter group based on the third indicator values ​​of the multiple third parameter groups; and updating the recommendation model based on the at least one fourth parameter group.

[0049] Therefore, the parameters of the recommendation model can be updated more efficiently and accurately based on a multi-round progressive evolutionary process. Referring to the further parameter evolution process described above, after the second parameter set has been generated and validated, the system can continue to initiate the next iteration. That is, based on the user's interaction information with the second output results of the second parameter set, the respective second indicator values ​​are calculated. Then, based on the indicator to be optimized, the second parameter set, and its corresponding second indicator values, a third prompt word is constructed, and the large model is called again for review and evolution. In this case, the large model can further derive multiple associated third evolutionary strategies based on the second evolutionary strategy that has been validated, and generate multiple third parameter sets for validation based on these third evolutionary strategies. By applying multiple rounds of "generation-indicator evaluation-screening evolution" iterations to update the parameters of the recommendation model, more accurate parameter evolution strategy information can be accumulated in multiple evolutionary rounds to obtain higher-quality candidate parameter sets.

[0050] According to some embodiments, a plurality of third evolutionary strategies associated with at least one second evolutionary strategy are obtained by maintaining the evolutionary direction of at least one second evolutionary strategy and adjusting the evolutionary step size of at least one second evolutionary strategy.

[0051] According to the above implementation method, the evolutionary direction refers to the relative trend of parameter value adjustment in the parameter space, and the evolutionary step size refers to the magnitude of parameter value adjustment in different evolutionary directions. By applying the above methods to determine multiple third evolutionary strategies associated with at least one second evolutionary strategy, parameter optimization can be achieved more efficiently, improving the parameter update efficiency of the recommendation model. For example, if the large model determines that the second evolutionary strategy of "increasing the value of parameter A and decreasing the value of parameter B" is a successful evolutionary strategy in the previous round, the magnitude of the increase in the value of parameter A and the magnitude of the decrease in the value of parameter B can be further adjusted on this basis to achieve more refined local optimization.

[0052] In some examples, multiple third evolutionary strategies can also be implemented in other ways, such as by randomly perturbing the parameter update direction determined by the second evolutionary strategy to explore locally near the validated effective direction, or by exploring new evolutionary directions by fusing the evolutionary directions of different second evolutionary strategies when multiple second evolutionary strategies exist. This way, while inheriting successful evolutionary experience, the diversity of parameter optimization directions is maintained, the optimization efficiency is improved, the probability of getting trapped in local optima is reduced, and the accuracy of the recommendation model update is improved.

[0053] According to some embodiments, method 200 further includes: in response to determining that multiple parameter items have been updated multiple times, determining the update convergence state information of multiple parameter items based on the indicator value change information of the indicator to be optimized in multiple updates, wherein step S201, determining the first prompt word based on the indicator to be optimized for the recommendation model, the current parameter values ​​of multiple parameter items, and the current indicator value of the indicator to be optimized, includes: determining the first prompt word based on the indicator to be optimized, the current parameter values ​​of multiple parameter items, the current indicator value, and the update convergence state information, and wherein the large model determines the evolution step size of multiple first evolution strategies based on the update convergence state information to generate multiple first parameter sets. Thus, the convergence state of model parameter updates can be passed to the large model as prior knowledge, enabling the large model to more accurately determine the evolution step size, improving parameter optimization efficiency and convergence accuracy, and enhancing the accuracy of recommendation model updates.

[0054] In some examples, updating convergence state information is used to characterize the current stage of the optimization process, such as describing whether parameter updates have stabilized or whether there is still significant room for improvement. In one example, the system can plot the curve of the value of the target indicator changing with the update rounds, calculate the corresponding statistical indicators (such as moving average or standard deviation) of the target indicator values ​​in the most recent rounds, and then determine whether the current parameter update process is in a "converging" state, to characterize whether the current parameter value is close to the local optimum under the current strategy.

[0055] In some examples, when convergence state information is passed to the large model, the model can respond to this information by adaptively adjusting the evolutionary step size of each first evolutionary strategy, thereby generating multiple first parameter sets. For instance, when the parameter update process is already in a convergent state, the large model can use a smaller evolutionary step size to generate candidate parameter sets to avoid skipping the optimal region due to an excessively large step size. When the parameter update process is in a non-convergent state, the large model will use a larger step size to more quickly approach the better region, improving parameter optimization efficiency.

[0056] According to some embodiments, step S205, which involves using a second prompt word to invoke a large model so that the large model generates at least one second parameter set, includes: in response to determining that the first comparison result meets preset conditions, invoking the large model using the second prompt word so that the large model generates at least one second parameter set, and updating the recommendation model based on at least one second parameter set. The preset conditions include that the first indicator value corresponding to at least one first parameter set is better than the current indicator value. Furthermore, method 200 further includes: in response to determining that the first comparison result does not meet preset conditions, invoking the large model to generate at least one fifth parameter set based on the current parameter values ​​of multiple parameter items and a fifth evolutionary strategy different from multiple first evolutionary strategies, and updating the recommendation model based on at least one fifth parameter set. Thus, preset conditions can be used to screen the effectiveness of randomly generated evolutionary strategies. When a first indicator value better than the current indicator value exists, further evolution is performed based on the successfully explored effective evolutionary strategy. When none of the randomly generated evolutionary parameter sets are successfully explored, an evolutionary exploration in a new direction is initiated to improve the accuracy of parameter optimization.

[0057] According to the above implementation method, the large model determines the next operation based on whether the comparison results between the first indicator values ​​of multiple first parameter groups and the current indicator value meet a preset condition. The preset condition is set to at least one first indicator value of a first parameter group is better than the current indicator value, that is, this round of exploration has generated at least one set of parameter configurations that can bring positive benefits. When the preset condition is met, it indicates that there is an effective strategy among the multiple first evolutionary strategies randomly generated by the large model, and the large model can be called to continue optimizing in the verified effective direction to generate at least one second parameter group. When the preset condition is not met, that is, the performance of all first parameter groups fails to surpass the current configuration, it indicates that the multiple randomly generated first evolutionary strategies are not advisable, and the above-mentioned rollback mechanism can be triggered, calling the large model to regenerate a fifth evolutionary strategy that is different from all the first evolutionary strategies verified as invalid based on the original current parameter values, so as to realize the rapid adjustment of the parameter evolution direction, improve the parameter optimization efficiency, and improve the accuracy of parameter optimization.

[0058] According to some embodiments, step S205, which involves using a second prompt word to invoke the large model so that the large model generates at least one second parameter set, includes: invoking the large model using the second prompt word so that the large model generates at least one second parameter set based on a first comparison result between the corresponding first indicator values ​​of multiple first parameter sets and the current indicator values, and a second comparison result between the multiple first parameter sets and the current parameter values ​​of multiple parameter items. By applying the above method, parameter adjustment strategies that can improve model performance indicators can be analyzed more accurately based on the comparison of the superiority and inferiority of indicator values ​​and the information on changes in parameter values, thereby generating optimized second parameter sets more accurately and improving the accuracy of parameter optimization.

[0059] In some examples, method 200 is used to optimize the parameters of a reordering model in a recommender system. The reordering model outputs a sequence of recommended content including multiple content blocks based on multiple content blocks and multiple sets of prediction information corresponding to each content block. Accordingly, according to some embodiments, the recommender model outputs a sequence of recommended content including multiple content blocks based on multiple content blocks and multiple sets of prediction information corresponding to each content block, with multiple parameter terms including importance coefficients of the multiple sets of prediction information corresponding to each content block. This allows for more efficient and accurate optimization of the reordering model's parameters, thereby improving the efficiency and accuracy of recommender reordering.

[0060] In some examples, multiple content blocks refer to content units awaiting sorting and recommendation. In video feed recommendation scenarios, multiple content blocks can be multiple short videos or live streams; in e-commerce scenarios, multiple content blocks can be multiple product cards; in news and information scenarios, multiple content blocks can be multiple news article entries. Multiple sets of prediction information are obtained by predicting multiple user interaction metrics for each content block. In one example, multiple sets of prediction information can be determined by other deep learning models or algorithm modules. As mentioned earlier, the importance coefficient corresponds to the adjustable parameters in the re-ranking model corresponding to prediction information of different dimensions, determining the contribution weight of each prediction information in the fused ranking score. In one example, the importance coefficient can be a weighted coefficient, where the model directly performs a linear weighted sum of each prediction information, and the overall ranking score of each content block is equal to the sum of the products of each prediction score and its corresponding weighted coefficient. In another example, the importance coefficient can also be a power coefficient, where the model first performs an exponential transformation on each prediction score before multiplying them, adjusting the influence of each set of prediction information in the final ranking process in a non-linear manner, thereby improving the accuracy of the recommended content sequence.

[0061] In some examples, the method 200 described above is used to optimize the parameters of the reordering model for various scenarios or user groups. For instance, in practical applications, different importance coefficients can be applied to content block reordering for different scenarios (e.g., mobile or web) or different user groups (e.g., light or heavy users). For example, when the number of scenario categories is 'a', the number of user groups is 'b', and the number of parameters is 'c', the evolutionary agent needs to learn 'a' × 'b' × 'c' parameters. In this case, the above-described metric evaluation steps are implemented based on different scenarios or different user groups, thereby achieving multi-set parameter optimization to improve recommendation accuracy.

[0062] In one example, when the content block is a video, the predicted information may include: estimated duration score, representing the expected duration a user is likely to watch the video; estimated completion score, representing the probability that a user will watch the video completely; estimated satisfaction duration score, representing the expected duration a user will watch in a satisfied state; 1-estimated quick swipe score, where the quick swipe score is the probability that a user will quickly swipe past the video, and its complement is used to positively represent the user's satisfaction level; completion rate, i.e., the proportion of the user's actual viewing time to the total video duration; and estimated interaction, representing the probability that the user will perform interactive operations such as liking, commenting, and sharing the video. By flexibly combining the above multiple sets of predicted information based on importance coefficients, the recommendation model can achieve fine-grained control among multiple business objectives. By applying the above method 200 to efficiently and automatically optimize multiple importance coefficients, the iteration efficiency and accuracy of the re-ranking model can be improved, thereby improving the recommendation accuracy.

[0063] According to some embodiments, the multiple sets of predictive information corresponding to each content block include the probability that a user browses the content block, the duration of the user's browsing of the content block, the type of interaction the user has with the content block, the probability that the user performs a satisfactory interaction with the content block, or multiple factors related to the content block. Therefore, multiple sets of predictive information can be used to more accurately and comprehensively characterize the user's satisfaction with the recommended content. By flexibly combining the above-mentioned multiple sets of predictive information based on importance coefficients, the recommendation model can achieve fine-grained control among multiple business objectives, thereby improving recommendation accuracy.

[0064] In some examples, the probability of a user browsing a content block is used to characterize whether the user will click on it to view it; the browsing duration of a user browsing a content block is used to characterize the length of time a user might stay while viewing it; the type of interaction a user interacts with a content block is used to characterize whether the user simply browses or engages in deep interaction (e.g., based on likes, comments, etc.); and the probability of a user performing a satisfying interaction with a content block is used to more intuitively indicate the probability of the user being satisfied with the content block. Satisfactory interactions can include, for example, likes, favorites, positive ratings, and watching the entire content. In addition to the above-listed items, predictive information can also include other types, such as the probability of a user sharing or forwarding the content block, and the probability of a user giving negative feedback (e.g., dislikes, reports, disinterest, etc.) to more accurately and comprehensively characterize the user's satisfaction with the recommended content. Furthermore, by flexibly combining multiple sets of predictive information, the accuracy of recommendations can be improved.

[0065] Figure 3 A schematic diagram illustrating the update process of a large-model-based recommendation model according to an exemplary embodiment of this disclosure is shown. See also Figure 3In this example, the update process of the recommendation model is collaboratively completed by the prompt word generation module 301, the large model 302, and the feedback evaluation module 303. First, the prompt word generation module 301 generates prompt words for calling the large model based on the optimization goal and evolutionary data. The optimization goal indicates the expected direction of business metric improvement in the update rounds of the recommendation model, such as "the optimization goal is to increase the total browsing time of users." Evolutionary data refers to the current model parameters or performance data, or further includes evolutionary experience information accumulated from previous update rounds, such as the current parameter values ​​of the model, candidate parameter groups generated in previous rounds, historically selected update parameter groups, and metric evaluation information obtained from user interaction verification for each parameter group. In one example, the prompt word generation module 301 concatenates and organizes the optimization goal and evolutionary data according to a preset template to obtain parameter optimization task instructions for calling the large model. The large model 302 generates candidate parameter groups for the recommendation model based on the prompt words. The feedback evaluation module 303 is used to configure the candidate parameter sets output by the large model 302 into the recommendation model and perform small-scale validation. It collects and statistically analyzes real user interaction data on the recommendation model's output results, and then calculates the evaluation information corresponding to each candidate parameter set, such as the average viewing time or interaction rate per user under each parameter set. The evaluation information and the aforementioned candidate parameter sets produced by the feedback evaluation module 303 can then be used as new evolutionary information for the next round of optimization iteration. This information is passed to the prompt word generation module 301 as input for the next iteration, using a multi-round "generation-evaluation-filtering evolution" iteration to update the recommendation model's parameters. This allows for the accumulation of more accurate parameter evolution strategy information across multiple evolutionary rounds, resulting in higher-quality candidate parameter sets. In some examples, when the candidate parameter sets output by the large model 302 are generated based on validated evolutionary experience, the recommendation model can be updated based on the current output parameter sets to obtain a more accurate recommendation model and improve the accuracy of online content recommendations.

[0066] According to one aspect of this disclosure, an update device for a recommendation model based on a large model is also provided. Figure 4 A structural block diagram of an update apparatus 400 for a recommendation model based on a large model, according to an exemplary embodiment of the present disclosure, is shown. Figure 4 The device 400 includes: The first determining unit 401 is configured to determine the first prompt word based on the indicator to be optimized for the recommendation model, the current parameter values ​​of multiple parameter items to be updated included in the recommendation model, and the current indicator value of the indicator to be optimized, wherein the indicator to be optimized includes an indicator that characterizes the user's interaction information with the output result of the recommendation model. The first calling unit 402 is configured to call the large model using the first prompt word, so that the large model generates multiple first parameter groups, wherein each of the multiple first parameter groups includes multiple candidate parameter values ​​corresponding to multiple parameter items respectively; The second determining unit 403 is configured to determine, for each of the plurality of first parameter groups, a first indicator value corresponding to the indicator to be optimized for that first parameter group based on the user's interaction information with the first output result corresponding to each of the plurality of first parameter groups, wherein the first output result is obtained using a recommendation model configured based on the first parameter group; The third determining unit 404 is configured to determine the second prompt word based on the first prompt word, multiple first parameter groups, and the corresponding first index values ​​of each of the multiple first parameter groups. The second invocation unit 405 is configured to invoke the large model using the second prompt word, such that the large model generates at least one second parameter set; and Update unit 406 is configured to update the recommendation model based on at least one second parameter group.

[0067] According to some embodiments, the first calling unit 402 is configured to call a large model using a first prompt word, such that the large model generates multiple first parameter groups based on multiple first evolution strategies, and wherein the large model performs the following operations based on a second prompt word: determining at least one second evolution strategy from multiple first evolution strategies based on a first comparison result of the first index value of each of the multiple first parameter groups with the current index value; and generating at least one second parameter group based on at least one second evolution strategy.

[0068] According to some embodiments, the apparatus 400 further includes: a fourth determining unit configured to determine a second indicator value corresponding to the indicator to be optimized for each of the at least one second parameter group based on user interaction information of the second output results corresponding to the at least one second parameter group; a fifth determining unit configured to determine a third prompt word based on the indicator to be optimized, the at least one second parameter group, and the second indicator value of each of the at least one second parameter group; a third invocation unit configured to invoke a large model using the third prompt word, such that the large model generates multiple third parameter groups based on multiple third evolutionary strategies associated with the second evolutionary strategy; a sixth determining unit configured to determine a third indicator value corresponding to the indicator to be optimized for each third parameter group based on user interaction information of the third output results corresponding to each third parameter group; a seventh determining unit configured to determine at least one fourth parameter group based on the third indicator values ​​corresponding to each of the multiple third parameter groups; and an updating unit further configured to update the recommendation model based on at least one fourth parameter group.

[0069] According to some embodiments, a plurality of third evolutionary strategies associated with at least one second evolutionary strategy are obtained by maintaining the evolutionary direction of at least one second evolutionary strategy and adjusting the evolutionary step size of at least one second evolutionary strategy.

[0070] According to some embodiments, the apparatus 400 further includes: an eighth determining unit, configured to determine update convergence state information of multiple parameter items based on the index value change information of the index to be optimized in the multiple rounds of updates in response to determining that multiple parameter items have been updated multiple times, wherein the first determining unit 401 is configured to determine a first prompt word based on the index to be optimized, the current parameter value of multiple parameter items, the current index value and the update convergence state information, and wherein the large model determines the evolution step size of multiple first evolution strategies based on the update convergence state information to generate multiple first parameter sets.

[0071] According to some embodiments, the second invocation unit 405 is configured to, in response to determining that the first comparison result meets preset conditions, invoke a large model using a second prompt word, so that the large model generates at least one second parameter group and updates the recommendation model based on at least one second parameter group. The preset conditions include that the first index value corresponding to at least one first parameter group is better than the current index value. Furthermore, the device 400 further includes: a fourth invocation unit, configured to, in response to determining that the first comparison result does not meet preset conditions, invoke a large model to generate at least one fifth parameter group based on the current parameter values ​​of multiple parameter items and a fifth evolutionary strategy different from multiple first evolutionary strategies, so as to update the recommendation model based on at least one fifth parameter group.

[0072] According to some embodiments, the second invocation unit 405 is configured to invoke the large model using a second prompt word, such that the large model generates at least one second parameter group based on the first comparison result of the first indicator value of each of the multiple first parameter groups with the current indicator value and the second comparison result of the multiple first parameter groups with the current parameter value of the multiple parameter items.

[0073] According to some embodiments, the recommendation model is used to output a sequence of recommended content including multiple content blocks based on multiple content blocks and multiple sets of prediction information corresponding to each content block, and multiple parameter terms include the importance coefficient of the multiple sets of prediction information corresponding to each content block.

[0074] According to some embodiments, the multiple sets of prediction information corresponding to each content block include the probability that a user browses the content block, the duration of the user's browsing of the content block, the type of interaction the user has with the content block, the probability that the user performs a satisfactory interaction with the content block, or multiple of the above.

[0075] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0076] According to another aspect of this disclosure, an electronic device is also provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method for updating the recommendation model based on a large model.

[0077] According to another aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is also provided, wherein the computer instructions are used to cause the computer to perform the above-described method for updating the recommendation model based on a large model.

[0078] According to another aspect of this disclosure, a computer program product is also provided, comprising a computer program, wherein the computer program, when executed by a processor, implements the above-described method for updating the recommendation model based on a large model.

[0079] refer to Figure 5 The present invention describes a structural block diagram of an electronic device 500 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0080] like Figure 5 As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 502 or a computer program loaded from storage unit 508 into random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. Input / output (I / O) interface 505 is also connected to bus 504.

[0081] Multiple components in device 500 are connected to I / O interface 505, including: input unit 506, output unit 507, storage unit 508, and communication unit 509. Input unit 506 can be any type of device capable of inputting information to device 500. Input unit 506 can receive input numerical or character information and generate key signal inputs related to user settings and / or function control of the electronic device, and may include, but is not limited to, a mouse, keyboard, touchscreen, trackpad, trackball, joystick, microphone, and / or remote control. Output unit 507 can be any type of device capable of presenting information, and may include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 508 may include, but is not limited to, a hard disk and an optical disk. Communication unit 509 allows device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth™ devices, 802.11 devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0082] The computing unit 501 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as the method for updating a recommendation model based on a large model. For example, in some embodiments, the method for updating a recommendation model based on a large model can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of the method for updating a recommendation model based on a large model described above can be performed. Alternatively, in other embodiments, computing unit 501 may be configured by any other suitable means (e.g., by means of firmware) to perform an update method for a recommendation model based on a large model.

[0083] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0084] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0085] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0086] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0087] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

[0088] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0089] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0090] While embodiments or examples of this disclosure have been described with reference to the accompanying drawings, it should be understood that the methods, systems, and devices described above are merely exemplary embodiments or examples, and the scope of the invention is not limited by these embodiments or examples, but only by the granted claims and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. Furthermore, the steps may be performed in a different order than that described in this disclosure. Further, various elements in the embodiments or examples may be combined in various ways. Importantly, as the technology evolves, many elements described herein can be replaced by equivalents that appear after this disclosure.

Claims

1. A method of updating a large model-based recommendation model, wherein, The recommendation model includes multiple parameter items to be updated, and the method includes: The first prompt word is determined based on the indicator to be optimized for the recommendation model, the current parameter values ​​of the plurality of parameter items, and the current indicator value of the indicator to be optimized, wherein the indicator to be optimized includes an indicator that characterizes the user's interaction information with the output result of the recommendation model; The first prompt word is used to invoke a large model so that the large model generates multiple first parameter groups, wherein each of the multiple first parameter groups includes multiple candidate parameter values ​​corresponding to the multiple parameter items respectively; For each of the plurality of first parameter groups, based on the user's interaction information with the first output result corresponding to each of the plurality of first parameter groups, a first indicator value corresponding to the indicator to be optimized for that first parameter group is determined, wherein the first output result is obtained using the recommendation model configured based on that first parameter group; The second prompt word is determined based on the first prompt word, the plurality of first parameter groups, and the corresponding first index value of each of the plurality of first parameter groups; The large model is invoked using the second prompt word, causing the large model to generate at least one second parameter set; and The recommendation model is updated based on at least one set of second parameters.

2. The method of claim 1, wherein, The step of using the first prompt word to invoke the large model, so that the large model generates multiple first parameter sets, includes: The first prompt word is used to invoke a large model, so that the large model generates the multiple first parameter sets based on multiple first evolutionary strategies respectively. Furthermore, the large model performs the following operations based on the second prompt word: Based on the first comparison result between the corresponding first index value of each of the plurality of first parameter groups and the current index value, at least one second evolutionary strategy is determined from the plurality of first evolutionary strategies; and The at least one second parameter set is generated based on the at least one second evolutionary strategy.

3. The method of claim 2, further comprising: Based on the user's interaction information with the corresponding second output results of the at least one second parameter group, determine the second indicator value of each of the at least one second parameter group corresponding to the indicator to be optimized; The third prompt word is determined based on the indicator to be optimized, the at least one second parameter group, and the second indicator value of each of the at least one second parameter group; The large model is invoked using the third prompt word, so that the large model generates multiple third parameter sets based on multiple third evolutionary strategies associated with the second evolutionary strategy. Based on the user's interaction information with the third output result corresponding to each third parameter group, the third indicator value corresponding to the indicator to be optimized for that third parameter group is determined. Based on the corresponding third index values ​​of each of the plurality of third parameter groups, at least one fourth parameter group is determined; as well as The recommendation model is updated based on at least one fourth parameter group.

4. The method of claim 3, wherein, The plurality of third evolutionary strategies associated with the at least one second evolutionary strategy are obtained by maintaining the evolutionary direction of the at least one second evolutionary strategy and adjusting the evolutionary step size of the at least one second evolutionary strategy.

5. The method according to any one of claims 2-4, further comprising: In response to the determination that the plurality of parameter items have been updated multiple times, the update convergence status information of the plurality of parameter items is determined based on the indicator value change information of the indicator to be optimized in the multiple rounds of updates. The step of determining the first prompt word based on the indicator to be optimized for the recommendation model, the current parameter values ​​of the plurality of parameter items, and the current indicator value of the indicator to be optimized includes: The first prompt word is determined based on the indicator to be optimized, the current parameter values ​​of the multiple parameter items, the current indicator value, and the updated convergence status information. Furthermore, the large model determines the evolution step size of the plurality of first evolution strategies based on the updated convergence state information, so as to generate the plurality of first parameter sets.

6. The method of claim 2, wherein, The step of using the second prompt word to invoke the large model, so that the large model generates at least one second parameter set, includes: In response to determining that the first comparison result meets preset conditions, the large model is invoked using the second prompt word, so that the large model generates the at least one second parameter group, and the recommendation model is updated based on the at least one second parameter group. The preset conditions include that the first indicator value corresponding to at least one first parameter group is better than the current indicator value. Furthermore, the method further includes: In response to determining that the first comparison result does not meet the preset condition, the large model is invoked to generate at least one fifth parameter group based on the current parameter values ​​of the plurality of parameter items and a fifth evolutionary strategy different from the plurality of first evolutionary strategies, so as to update the recommendation model based on the at least one fifth parameter group.

7. The method of claim 1, wherein, The step of using the second prompt word to invoke the large model, so that the large model generates at least one second parameter set, includes: The large model is invoked using the second prompt word, so that the large model generates at least one second parameter group based on the first comparison result of the first indicator value of each of the plurality of first parameter groups with the current indicator value and the second comparison result of the plurality of first parameter groups with the current parameter value of the plurality of parameter items.

8. The method according to any one of claims 1-7, wherein, The recommendation model is used to output a recommended content sequence including the multiple content blocks based on multiple content blocks and multiple sets of prediction information corresponding to each content block. The multiple parameter items include the importance coefficient of the multiple sets of prediction information corresponding to each content block.

9. The method of claim 8, wherein, For each content block, there are multiple sets of prediction information, including the probability that a user will browse the content block, the duration of the user's browsing of the content block, the type of interaction the user will have with the content block, the probability that the user will perform a satisfactory interaction with the content block, or multiple factors related to the content block.

10. An update device for a recommendation model based on a large model, wherein, The recommendation model includes multiple parameter items to be updated, and the device includes: The first determining unit is configured to determine a first prompt word based on the indicator to be optimized for the recommendation model, the current parameter values ​​of the plurality of parameter items, and the current indicator value of the indicator to be optimized, wherein the indicator to be optimized includes an indicator characterizing the user's interaction information with the output result of the recommendation model. The first invocation unit is configured to invoke the large model using the first prompt word, so that the large model generates multiple first parameter groups, wherein each of the multiple first parameter groups includes multiple candidate parameter values ​​corresponding to the multiple parameter items respectively; The second determining unit is configured to, for each of the plurality of first parameter groups, determine a first indicator value corresponding to the indicator to be optimized for that first parameter group based on user interaction information of the first output result corresponding to each of the plurality of first parameter groups, wherein the first output result is obtained using the recommendation model configured based on the first parameter group; The third determining unit is configured to determine the second prompt word based on the first prompt word, the plurality of first parameter groups, and the corresponding first index value of each of the plurality of first parameter groups; The second invocation unit is configured to invoke the large model using the second prompt word, such that the large model generates at least one second parameter set; and The update unit is configured to update the recommendation model based on the at least one second parameter group.

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

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

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

Images