Method and device for determining to-be-recommended resource based on large model, equipment and storage medium

Abstract

The present disclosure provides a method and device for determining a recommended resource based on a large model, an equipment and a storage medium, which relates to the technical fields of deep learning, multimedia processing and the like. The method comprises: obtaining a behavior preference of a current user and a current refresh depth; determining a candidate resource and a similarity corresponding to the candidate resource according to the behavior preference and an initial resource; determining resource satisfaction information of the current user at the current refresh depth according to the current refresh depth, a historical resource played by the current user and playing attribute information of the historical resource, wherein the playing attribute information at least comprises a historical refresh depth; screening the candidate resource according to the resource satisfaction information and the similarity, and determining a recommended resource of the current user at the current refresh depth based on a large model according to the screened candidate resource.

Description

Technical Field

[0001] This disclosure relates to the field of computer science, and more particularly to the field of artificial intelligence, including deep learning, multimedia processing, and other related technologies. Specifically, it relates to a method, apparatus, device, and storage medium for determining resources to be recommended based on a large model. Background Technology

[0002] With the rapid advancement of information technology and the widespread application of big data and artificial intelligence, recommendation systems are booming at an unprecedented pace, especially demonstrating enormous potential and vitality in short video applications. Users only need to simply input their behavioral preferences or conduct initial exploration based on personal interests, and the recommendation system can instantly analyze massive amounts of data to accurately push short video content that matches their tastes. While enjoying the pleasant experience brought by personalized recommendations, users can also seamlessly transition to the next exciting video by constantly refreshing the page, achieving a smooth continuous playback experience. How to recommend resources in continuous playback scenarios is crucial. Summary of the Invention

[0003] This disclosure provides a method, apparatus, device, and storage medium for determining resources to be recommended based on a large model.

[0004] According to one aspect of this disclosure, a method for determining resources to be recommended based on a large model is provided, comprising:

[0005] Get the current user's behavioral preferences and current refresh depth;

[0006] Based on the behavioral preferences and initial resources, determine the similarity between the candidate resources and the corresponding similarities.

[0007] Based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources, the resource satisfaction information of the current user at the current refresh depth is determined; wherein, the playback attribute information includes at least the historical refresh depth;

[0008] The candidate resources are filtered based on the resource satisfaction information and the similarity, and the recommended resources for the current user at the current refresh depth are determined based on the filtered candidate resources using a large model.

[0009] According to one aspect of this disclosure, an apparatus for determining resources to be recommended based on a large model is provided, comprising:

[0010] The current acquisition module is used to acquire the current user's behavioral preferences and current refresh depth;

[0011] The candidate resource module is used to determine the similarity between candidate resources and the candidate resources based on the behavioral preferences and the initial resources.

[0012] The resource satisfaction module is used to determine the current user's resource satisfaction information at the current refresh depth based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources; wherein, the playback attribute information includes at least the historical refresh depth;

[0013] The recommended resource module is used to filter the candidate resources based on the resource satisfaction information and the similarity, and determine the recommended resources for the current user at the current refresh depth based on the filtered candidate resources using a large model.

[0014] According to another aspect of this disclosure, an electronic device is provided, the electronic device comprising:

[0015] At least one processor; and

[0016] A memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods provided in any embodiment of this disclosure.

[0018] 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 a computer to perform the methods provided in any embodiment of this disclosure.

[0019] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods provided according to any embodiment of this disclosure.

[0020] 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

[0021] Figure 1 This is an architecture diagram of a remote logging system provided according to an embodiment of the present disclosure;

[0022] Figure 2 This is a flowchart of another method for determining resources to be recommended based on a large model, according to an embodiment of this disclosure;

[0023] Figure 3 This is a flowchart of another method for determining resources to be recommended based on a large model, according to an embodiment of this disclosure;

[0024] Figure 4This is a schematic diagram of a device for determining resources to be recommended based on a large model, according to an embodiment of this disclosure.

[0025] Figure 5 This is a block diagram of an electronic device used to implement the method for determining resources to be recommended based on a large model according to embodiments of the present disclosure. Detailed Implementation

[0026] Figure 1 This is a flowchart illustrating a method for determining resources to be recommended based on a large model, according to embodiments of this disclosure. This method is applicable to situations where resources are pushed to users. The method can be executed by an apparatus for determining resources to be recommended based on a large model, which can be implemented in software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method for determining resources to be recommended based on a large model in this embodiment may include:

[0027] S101, obtain the current user's behavioral preferences and current refresh depth;

[0028] S102, Based on the behavioral preferences and initial resources, determine the similarity between the candidate resources and the corresponding similarity.

[0029] S103, determine the current user's resource satisfaction information at the current refresh depth based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources; wherein, the playback attribute information includes at least the historical refresh depth;

[0030] S104, the candidate resources are filtered according to the resource satisfaction information and the similarity, and the recommended resources for the current user at the current refresh depth are determined based on the filtered candidate resources using a large model.

[0031] The current user refers to the user who currently needs resource recommendations, that is, the user who needs to play the resource. The current user's behavioral preferences characterize their behavioral characteristics, interests, historical choices, or personal preferences. The current refresh depth characterizes the depth or sequence continuity of continuous playback within the current playback session, reflecting the level of continuity and depth of exploration during playback. The current playback session can be initiated when the user launches the resource application or enters a query request. If the user enters a query request via text or voice, that query request reflects the current user's behavioral characteristics.

[0032] For example, in the current playback session, resources can be pushed to the current user based on their behavioral preferences, the first resource played in the current playback session can be retrieved, and a refresh operation can be performed to seamlessly continue playing new resources. For the resource played by the current user for the i-th time (i is a positive integer) in the current playback session, the corresponding refresh depth is i, where i is a positive integer.

[0033] Historical resources refer to the resources played by the current user in the historical playback session. Historical refresh depth characterizes the depth or sequence continuity of continuous playback within the historical playback session. Historical playback sessions precede the current playback session. Resource satisfaction information characterizes the current user's satisfaction with resources at the current refresh depth, serving as a resource interest signal based on the current refresh depth. This information is used to more accurately and promptly grasp the current user's resource preferences at the current refresh depth.

[0034] For example, the current user's behavioral preferences can be encoded to obtain a behavioral preference vector representation (embedding). The semantic similarity between this behavioral preference vector representation and the vector representations of each initial resource can be determined. Multiple initial resources with relatively high semantic similarity can be selected as candidate resources, and the similarity of these candidate resources can be calculated. The historical resources played by the current user in the historical playback session and the historical refresh depth of these historical resources can be obtained, resulting in playback attribute information that includes at least the historical refresh depth. The playback attribute information of historical resources may also include the viewing duration, completion rate, or interaction behavior of the historical resources.

[0035] For example, the user's satisfaction with historical resources can be determined based on their playback attribute information, thus obtaining the user's satisfaction information with historical resources. The user's satisfaction with resources at the current refresh depth can then be determined by combining the current refresh depth, the depth of historical resources, and the satisfaction information with historical resources. Furthermore, the user can use the resource satisfaction information at the current refresh depth and the similarity between candidate resources to filter them. Based on a large model, recommended resources can be determined for the current user at the current refresh depth from the remaining candidate resources after filtering. For instance, a large language model can be used to rank the filtered candidate resources, and the recommended resources can be determined based on the ranking results for the current user to play at the current refresh depth.

[0036] In resource recommendation scenarios, users gradually experience fatigue with increasing refresh depth. Specifically, as refresh depth increases, resource completion rate, effective playback rate, and playback duration all show a decreasing trend, while the rate of quick scrolling increases. Therefore, users' interests differ at different refresh depths. This embodiment of the disclosure determines resource satisfaction information by accurately fusing the current refresh depth, historical resources, and playback attribute information of historical resources. The process of determining resource satisfaction information not only considers the current refresh depth in the user's immediate behavior but also takes into account the user's long-term preferences and habits through the playback attribute information of historical resources, enabling a more accurate acquisition of the current user's resource satisfaction information at the current refresh depth. Furthermore, by introducing the resource satisfaction information at the current refresh depth as an interest signal into the resource recommendation process, and determining the resources to be recommended for the current user at the current refresh depth based on the resource satisfaction information and the similarity of each candidate resource, the matching degree between the recommended resources and the current user's needs is improved, thereby effectively enhancing the user experience.

[0037] The technical solution provided in this disclosure obtains the current refresh depth of the current user in the current playback session, the historical resources and their playback attribute information in the historical playback session, and combines the current refresh depth, historical resources and their playback attribute information to determine the current user's resource satisfaction information at the current refresh depth. The resource satisfaction information at the current refresh depth is then introduced as an interest signal into the resource recommendation process, thereby improving the matching degree between the recommended resources and the current user's needs and effectively enhancing the user experience.

[0038] Figure 2 This is a flowchart of another method for determining resources to be recommended based on a large model, according to embodiments of this disclosure. See also... Figure 2 Based on the above embodiments, the method for determining resources to be recommended based on a large model in this embodiment may include:

[0039] S201, obtain the current user's behavioral preferences and current refresh depth;

[0040] S202, Based on the behavioral preferences and initial resources, determine the similarity between the candidate resources and the corresponding similarities;

[0041] S203, based on the current refresh depth and the historical refresh depth of the historical resources, determine the reference resource associated with the current refresh depth from the historical resources;

[0042] S204, Based on the current user's playback duration of the reference resource and the total duration of the reference resource, determine the current user's resource satisfaction information at the current refresh depth;

[0043] S205, the candidate resources are filtered based on the resource satisfaction information and the similarity, and the recommended resources for the current user at the current refresh depth are determined based on the filtered candidate resources using a large model.

[0044] Historical refresh depth, playback duration, and total duration are playback attribute information for historical resources. During the filtering of candidate resources, reference resources associated with the current refresh depth can be selected from historical resources based on the current refresh depth and the historical refresh depth of the historical resources. For example, historical resources with the same historical refresh depth as the current refresh depth can be used as reference resources for the current refresh depth; alternatively, refresh depths can be grouped into multiple refresh depth intervals, and historical resources belonging to the same refresh depth interval as the current refresh depth can be used as reference resources for the current refresh depth. By selecting reference resources associated with the current refresh depth from historical resources, the matching degree between the reference resources and the current refresh depth is improved compared to using all historical resources as reference resources.

[0045] For example, the current user's satisfaction level with the reference resource can be determined based on the current user's playback duration and the total duration of the reference resource, and the resource satisfaction information at the current refresh depth can be determined based on the reference resource's satisfaction level. By using the playback duration and total duration of the reference resource to determine the resource satisfaction information at the current refresh depth, the matching degree between the resource satisfaction information and the current refresh depth is improved, that is, the accuracy of the resource satisfaction information is improved.

[0046] In one optional implementation, determining the current user's resource satisfaction information at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource includes: determining the current user's satisfaction with the reference resource at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource; selecting a first resource that the current user is satisfied with and a second resource that the current user is not satisfied with from the reference resource based on the satisfaction level; and determining the resource satisfaction information based on the first resource and the second resource.

[0047] For example, the user's satisfaction with the reference resources at the current refresh depth can be determined by combining the playback duration and total duration of the reference resources. The first set of reference resources with relatively high satisfaction levels are selected as the user's first satisfied resources; the second set of reference resources with relatively low satisfaction levels are selected as the user's second satisfied resources. The first and second sets are preset, for example, both can be 20. Resource satisfaction information is determined by combining the first and second resources, for example, by analyzing the differences in content, functionality, and other attributes between the first and second resources. By determining the user's satisfaction with each reference resource and selecting the user's satisfied first and dissatisfied second resources from the reference resources according to satisfaction level, the differences in content between the first and second resources can be fully considered, thus helping to improve the accuracy and quality of resource satisfaction information.

[0048] In one optional implementation, determining the current user's satisfaction with the reference resource at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource includes: determining the playback completion rate of the reference resource based on the current user's reference playback duration of the reference resource and the total duration of the reference resource; obtaining the maximum playback duration of a single resource from the reference playback duration, and determining the contribution of the reference resource based on the reference playback duration and the maximum playback duration; and determining the satisfaction rate based on the playback completion rate and the contribution.

[0049] For example, the playback completion rate of the reference resource can be obtained by dividing the reference playback duration by the total duration of the reference resource; alternatively, the playback completion rate of the reference resource can be determined using the following formula.

[0050] finish_rate=min(1.0, play_duration / sv_duration)

[0051] Among them, play_duration, sv_duration, and finish_rate are the reference playback duration, total duration, and playback completion rate, respectively, and min() is used to obtain the minimum value.

[0052] Furthermore, the maximum playback duration of a single resource is obtained from each reference playback duration. The contribution of the reference resource can be obtained by dividing the reference playback duration by the maximum playback duration. The satisfaction of the reference resource can be determined by combining the playback completion rate and the contribution of the reference resource. For example, the satisfaction rate can be positively correlated with the playback completion rate. The satisfaction rate can change rapidly and approach a fixed value as the contribution rate increases. The fixed value can be 1.

[0053] Users have different consumption habits. Some users may prefer to scroll quickly and rarely watch resources for extended periods, while others may watch most resources for a long time. By using the current user's maximum playback time as a benchmark, and determining the contribution of reference resources based on their reference playback time and maximum playback time, the matching degree between contribution and the current user can be improved, thus increasing the accuracy of contribution assessment. By combining the playback completion rate and contribution of reference resources to determine their satisfaction level, the influence of user consumption habits on satisfaction can be eliminated, thereby improving the accuracy of satisfaction assessment.

[0054] In one optional implementation, determining the satisfaction level based on the playback completion rate and the contribution rate includes determining the satisfaction level using the following formula:

[0055]

[0056] Here, `satisfy_score` represents satisfaction level, `finish_rate` represents playback completion rate, `play_rate` represents contribution rate, and `a` is a constant. For example, `a` can be 8. The above processing ensures a positive correlation between satisfaction level and playback completion rate, with satisfaction level increasing rapidly with increasing contribution rate until it stabilizes.

[0057] The technical solution provided in this disclosure determines the playback completion rate and contribution of reference resources by using the playback duration and total duration of reference resources, thereby determining the satisfaction level of reference resources; and selects each first resource that the current user is satisfied with and each second resource that the current user is dissatisfied with from the reference resources according to the satisfaction level; combining the first resources that the current user is satisfied with and the second resources that the current user is dissatisfied with to determine resource satisfaction information can fully consider the differences in resource content between the first resources and the second resources themselves, thereby helping to improve the accuracy and quality of resource satisfaction information.

[0058] In one optional implementation, determining the resource satisfaction information based on the first resource and the second resource includes: averaging the vector representations of each of the first resources to obtain a first resource vector, and averaging the vector representations of each of the second resources to obtain a second resource vector; determining a first correlation between the vector representation of the candidate resource and the first resource vector, and determining a second correlation between the vector representation of the candidate resource and the second resource vector; subtracting the second correlation from the first correlation to obtain the resource satisfaction coefficient for the current refresh depth; the resource satisfaction information is the resource satisfaction coefficient.

[0059] Taking the acquisition of 20 first resources and 20 second resources as an example, the average of the embeddings of the 20 first resources can be used to obtain a first resource vector, matching it with the current user's interests. Similarly, the average of the embeddings of the 20 second resources can be used to obtain a second resource vector, matching it with the current user's dislikes. Furthermore, a first correlation between the vector representations of candidate resources and the first resource vectors is determined, as well as a second correlation between the vector representations of candidate resources and the second resource vectors. The first correlation is subtracted from the second correlation to obtain the resource satisfaction coefficient at the current refresh depth, which is then used as resource satisfaction information. By subtracting the second correlation between the first resource vectors matching the interests and the second resource vectors corresponding to the dislikes, the resource satisfaction coefficient is obtained. This means the resource satisfaction coefficient can balance the current user's interests and dislikes, ensuring that the resources to be recommended, determined based on the resource satisfaction coefficient, not only match the current user's interests but also avoid their dislikes, further improving the matching degree between the recommended resources and the current user, thus enhancing the user experience.

[0060] In one optional implementation, the step of filtering the candidate resources based on the resource satisfaction information and the similarity includes: multiplying the resource satisfaction coefficient by the similarity to obtain a new similarity for the candidate resources; and filtering the candidate resources based on the new similarity.

[0061] For example, candidate resources with relatively low new similarity can be filtered out, and only candidate resources with relatively high new similarity can be retained. By multiplying the similarity of the candidate resources by the resource satisfaction coefficient to obtain the new similarity of the candidate resources, the influence of the resource satisfaction coefficient on the similarity can be retained when the resource satisfaction coefficient is much smaller than the similarity coefficient and is not zero, thereby reflecting the matching degree of the candidate resources.

[0062] Figure 3 This is a flowchart illustrating yet another method for determining resources to be recommended based on a large model, according to embodiments of this disclosure. See also... Figure 3 The method for determining resources to be recommended based on a large model in this embodiment may include:

[0063] S301, obtain the current user's behavioral preferences and current refresh depth;

[0064] S302, Based on the behavioral preferences and initial resources, determine the similarity between the candidate resources and the corresponding similarities;

[0065] S303, obtain the historical resources currently played by the user, and determine the number of historical resources currently played by the user at each historical refresh depth based on the historical refresh depth of the historical resources;

[0066] S304, the historical refresh depth is divided according to the number of historical resources to obtain at least two refresh depth intervals;

[0067] S305, determine the current refresh depth interval to which the current refresh depth belongs from the at least two refresh depth intervals, and use historical resources whose historical refresh depth belongs to the current refresh depth interval as reference resources;

[0068] S306, Based on the current user's playback duration of the reference resource and the total duration of the reference resource, determine the current user's resource satisfaction information at the current refresh depth;

[0069] S307, the candidate resources are filtered based on the resource satisfaction information and the similarity, and the recommended resources for the current user at the current refresh depth are determined based on the filtered candidate resources using a large model.

[0070] Among them, historical refresh depth, playback duration, and total duration are playback attribute information of historical resources.

[0071] During the screening of candidate resources, the system obtains the historical resources currently played by the user and the historical refresh depth of each historical resource in its corresponding historical playback session. For each historical refresh depth, it obtains the number of historical resources played by the current user at each historical refresh depth, i.e., the number of historical resources belonging to that historical refresh depth. For example, if the current user played 5 historical resources in the first historical playback session, 4 historical resources in the second historical playback session, and 7 historical resources in the third historical playback session, then the number of historical resources at the first, second, third, and fourth historical refresh depths is 3 each, the number of historical resources at the fifth historical refresh depth is 2, and the number of historical resources at the sixth and seventh historical refresh depths is 1 each.

[0072] For example, the system obtains the historical resources played by the current user in different historical playback sessions and the historical refresh depth of each historical resource. For each historical refresh depth, it obtains the number of historical playback sessions including that historical refresh depth, which is used as the number of historical resources for that historical refresh depth. Each historical refresh depth is divided according to the number of historical resources, resulting in at least two refresh depth intervals. The refresh depth interval to which the current refresh depth belongs is designated as the current refresh depth interval, and each historical resource belonging to the current refresh depth interval is used as a reference resource. By dividing each historical refresh depth using the number of historical resources to obtain at least two refresh depth intervals, using the refresh depth interval to which the current refresh depth belongs as the current depth interval, and using historical resources belonging to the current refresh depth interval as reference resources, the number of reference resources can be enriched, thereby improving the quality of subsequent processing based on the playback duration and total duration of the reference resources.

[0073] In one alternative implementation, the deviation between the number of historical resources in different refresh depth intervals is less than a preset deviation threshold.

[0074] For example, in the process of dividing each historical refresh depth into at least two refresh depth intervals, the number of historical resources in each refresh depth interval is made approximately equal. Since shallower historical refresh depths correspond to a relatively larger number of historical resources, and deeper historical refresh depths correspond to a relatively smaller number of historical resources, the process of dividing the historical refresh depths ensures that the number of historical resources corresponding to each refresh depth interval is approximately equal. Therefore, the number of historical refresh depths in a relatively shallow refresh depth interval is less than the number of historical refresh depths in a relatively deep refresh depth interval. For example, the first refresh depth interval may include 4 historical refresh depths (1-4), the second refresh depth interval may include 6 historical refresh depths (5-10), and the third refresh depth interval may include 10 historical refresh depths (11-20). By controlling the number of historical resources in each refresh depth interval to be approximately equal, the number of reference resources corresponding to each refresh depth interval is made approximately equal, thereby balancing the impact of reference resources on resource satisfaction information in each refresh depth interval and improving the generalization of resource depth information.

[0075] The technical solution provided in this disclosure divides each historical refresh depth into at least two refresh depth intervals by using the number of historical resources. The refresh depth interval to which the current refresh depth belongs is taken as the current depth interval, and the historical resources whose historical refresh depth belongs to the current refresh depth interval are taken as reference resources. This can enrich the number of reference resources, thereby improving the quality of subsequent processing based on the playback duration and total duration of the reference resources.

[0076] Figure 4This is a schematic diagram of a device for determining resources to be recommended based on a large model, according to an embodiment of this disclosure. This device is suitable for situations where resources are pushed to users. The device can be implemented in software and / or hardware and can be integrated into electronic devices. Figure 4 As shown, the apparatus 400 for determining resources to be recommended based on a large model in this embodiment may include:

[0077] The current acquisition module 410 is used to acquire the current user's behavioral preferences and current refresh depth;

[0078] The candidate resource module 420 is used to determine the similarity between candidate resources and the candidate resources based on the behavioral preferences and the initial resources.

[0079] The resource satisfaction module 430 is used to determine the current user's resource satisfaction information at the current refresh depth based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources; wherein, the playback attribute information includes at least the historical refresh depth.

[0080] The resource recommendation module 440 is used to filter the candidate resources based on the resource satisfaction information and the similarity, and to determine the resource to be recommended for the current user at the current refresh depth based on the filtered candidate resources using a large model.

[0081] In one alternative implementation, the resource satisfaction module 430 includes:

[0082] The reference resource submodule is used to determine the reference resource associated with the current refresh depth from the historical resources based on the current refresh depth and the historical refresh depth of the historical resources;

[0083] The resource satisfaction submodule is used to determine the current user's resource satisfaction information at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource; wherein the playback duration and the total duration belong to the playback attribute information.

[0084] In one optional implementation, the resource satisfaction submodule includes:

[0085] A satisfaction unit is used to determine the current user's satisfaction with the reference resource at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource.

[0086] The resource selection unit is configured to select, from the reference resources, a first resource that the current user is satisfied with and a second resource that the current user is not satisfied with, based on the satisfaction level.

[0087] The resource satisfaction unit is used to determine the resource satisfaction information based on the first resource and the second resource.

[0088] In one alternative implementation, the satisfaction unit includes:

[0089] The completion subunit is used to determine the playback completion degree of the reference resource based on the current user's reference playback duration of the reference resource and the total duration of the reference resource.

[0090] The contribution subunit is used to obtain the maximum playback duration of a single resource from the reference playback duration, and to determine the contribution of the reference resource based on the reference playback duration and the maximum playback duration.

[0091] The satisfaction subunit is used to determine the satisfaction level based on the playback completion rate and the contribution rate.

[0092] In one optional implementation, the satisfaction subunit is specifically used to determine the satisfaction level using the following formula:

[0093]

[0094] Where satisfaction_score is the satisfaction level, finish_rate is the completion rate, play_rate is the contribution rate, and 'a' is a constant.

[0095] In one optional implementation, the resource satisfaction unit includes:

[0096] The resource vector subunit is used to average the vector representations of each of the first resources to obtain a first resource vector, and to average the vector representations of each of the second resources to obtain a second resource vector.

[0097] The resource relevance subunit is used to determine a first relevance between the vector representation of the candidate resource and the first resource vector, and to determine a second relevance between the vector representation of the candidate resource and the second resource vector.

[0098] The satisfaction coefficient subunit is used to subtract the second relevance from the first relevance to obtain the resource satisfaction coefficient at the current refresh depth; the resource satisfaction information is the resource satisfaction coefficient.

[0099] In one alternative implementation, the resource recommendation module 440 includes:

[0100] The new similarity submodule is used to multiply the resource satisfaction coefficient by the similarity to obtain a new similarity for the candidate resource;

[0101] The resource filtering submodule is used to filter the candidate resources based on the new similarity.

[0102] In one alternative implementation, the reference resource submodule includes:

[0103] The resource quantity unit is used to determine the number of historical resources played by the current user at each of the historical refresh depths, based on the historical refresh depths of the historical resources.

[0104] A refresh depth division unit is used to divide each of the historical refresh depths using the number of historical resources, to obtain at least two refresh depth intervals.

[0105] A reference resource unit is used to determine the current refresh depth interval to which the current refresh depth belongs from the at least two refresh depth intervals; wherein, the reference resource is a historical resource whose historical refresh depth belongs to the current refresh depth interval.

[0106] In one alternative implementation, the deviation between the number of historical resources in different refresh depth intervals is less than a preset deviation threshold.

[0107] The apparatus for determining resources to be recommended based on a large model provided in this embodiment of the invention can execute the method for determining resources to be recommended based on a large model provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0108] The acquisition, storage, and application 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.

[0109] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0110] Figure 5 This is a block diagram of an electronic device used to implement the method for determining resources to be recommended based on a large model according to embodiments of the present disclosure. Figure 5 A schematic block diagram of an example electronic device 500 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0111] like Figure 5 As shown, the electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the electronic device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0112] Multiple components in electronic device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows electronic device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0113] The computing unit 501 can be a variety of 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 of determining resources to be recommended based on a large model. For example, in some embodiments, the method of determining resources to be recommended 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 the electronic 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 of determining resources to be recommended based on a large model described above can be performed. Alternatively, in other embodiments, computing unit 501 may be configured in any other suitable manner (e.g., by means of firmware) to perform a method for determining resources to be recommended based on a large model.

[0114] Figure 5A schematic block diagram of an example electronic device 500 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0115] like Figure 5 As shown, the electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the electronic device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0116] Multiple components in electronic device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows electronic device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0117] The computing unit 501 can be a variety of 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 of determining resources to be recommended based on a large model. For example, in some embodiments, the method of determining resources to be recommended 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 the electronic 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 of determining resources to be recommended based on a large model described above can be performed. Alternatively, in other embodiments, computing unit 501 may be configured in any other suitable manner (e.g., by means of firmware) to perform a method for determining resources to be recommended based on a large model.

[0118] 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.

[0119] 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.

[0120] 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.

[0121] 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, audio input, or tactile input).

[0122] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or network player through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0123] 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.

[0124] 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, audio recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0125] Cloud computing refers to a technology system that enables access to a shared pool of physical or virtual resources via a network. These resources can include servers, operating systems, networks, software, applications, and storage devices, and can be deployed and managed on demand and in a self-service manner. Cloud computing technology can provide efficient and powerful data processing capabilities for applications such as artificial intelligence and blockchain, as well as for model training.

[0126] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0127] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for determining resources to be recommended based on a large model, comprising: Obtain the current user's behavioral preferences and current refresh depth; wherein, the current refresh depth is used to characterize the degree of continuous playback depth or sequence continuity of the current user in the current playback session; Based on the behavioral preferences and initial resources, determine the similarity between the candidate resources and the corresponding similarities. Based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources, the resource satisfaction information of the current user at the current refresh depth is determined; wherein, the playback attribute information includes at least the historical refresh depth; The candidate resources are filtered based on the resource satisfaction information and the similarity, and the recommended resources for the current user at the current refresh depth are determined based on the filtered candidate resources using a large model. The step of determining the current user's resource satisfaction information at the current refresh depth based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources includes: Based on the current refresh depth and the historical refresh depth of historical resources, determine the reference resource associated with the current refresh depth from the historical resources; Based on the current user's playback duration of the reference resource and the total duration of the reference resource, the current user's satisfaction with the reference resource at the current refresh depth is determined; wherein, the playback duration and the total duration are part of the playback attribute information; Based on the satisfaction level, select a first resource that the current user is satisfied with and a second resource that the current user is not satisfied with from the reference resources; Based on the first resource and the second resource, determine the resource satisfaction information.

2. The method according to claim 1, wherein, The step of determining the current user's satisfaction with the reference resource at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource includes: The playback completion rate of the reference resource is determined based on the current user's reference playback duration and the total duration of the reference resource. Obtain the maximum playback duration of a single resource from the reference playback duration, and determine the contribution of the reference resource based on the reference playback duration and the maximum playback duration; The satisfaction level is determined based on the playback completion rate and the contribution level.

3. The method according to claim 2, wherein, The step of determining the satisfaction level based on the playback completion rate and the contribution rate includes determining the satisfaction level using the following formula: ； in, For satisfaction, For playback completion rate, As for contribution level, It is a constant.

4. The method according to any one of claims 1-3, wherein, The step of determining the resource satisfaction information based on the first resource and the second resource includes: The first resource vector is obtained by averaging the vector representations of each of the first resources, and the second resource vector is obtained by averaging the vector representations of each of the second resources. Determine a first correlation between the vector representation of the candidate resource and the first resource vector, and determine a second correlation between the vector representation of the candidate resource and the second resource vector; Subtracting the second relevance from the first relevance yields the resource satisfaction coefficient for the current refresh depth; the resource satisfaction information is the resource satisfaction coefficient.

5. The method according to claim 1, wherein, The step of filtering candidate resources based on the resource satisfaction information and the similarity includes: Multiply the resource satisfaction coefficient by the similarity to obtain a new similarity for the candidate resource; The candidate resources are filtered based on the new similarity.

6. The method according to claim 1, wherein, The step of determining a reference resource associated with the current refresh depth from the historical resources based on the current refresh depth and the historical refresh depth of the historical resources includes: Based on the historical refresh depth of the historical resources, determine the number of historical resources that the current user is playing at each of the historical refresh depths; The historical refresh depth is divided according to the number of historical resources to obtain at least two refresh depth intervals; The current refresh depth interval to which the current refresh depth belongs is determined from the at least two refresh depth intervals; wherein, the reference resource is a historical resource whose historical refresh depth belongs to the current refresh depth interval.

7. The method according to claim 6, wherein, The deviation between the number of historical resources in different refresh depth ranges is less than a preset deviation threshold.

8. An apparatus for determining resources to be recommended based on a large model, comprising: The current acquisition module is used to acquire the current user's behavioral preferences and current refresh depth; wherein, the current refresh depth is used to characterize the degree of continuous playback depth or sequence continuity of the current user in the current playback session; The candidate resource module is used to determine the similarity between candidate resources and the candidate resources based on the behavioral preferences and the initial resources. The resource satisfaction module is used to determine the current user's resource satisfaction information at the current refresh depth based on the current refresh depth, the historical resources played by the current user, and the playback attribute information of the historical resources; wherein, the playback attribute information includes at least the historical refresh depth; The resource recommendation module is used to filter the candidate resources based on the resource satisfaction information and the similarity, and determine the resource to be recommended for the current user at the current refresh depth based on the filtered candidate resources using a large model. The resource satisfaction module includes: The reference resource submodule is used to determine the reference resource associated with the current refresh depth from the historical resources based on the current refresh depth and the historical refresh depth of the historical resources; The resource satisfaction submodule is used to determine the current user's resource satisfaction information at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource; wherein, the playback duration and the total duration belong to the playback attribute information; The resource satisfaction submodule includes: A satisfaction unit is used to determine the current user's satisfaction with the reference resource at the current refresh depth based on the current user's playback duration of the reference resource and the total duration of the reference resource. The resource selection unit is configured to select, from the reference resources, a first resource that the current user is satisfied with and a second resource that the current user is not satisfied with, based on the satisfaction level. The resource satisfaction unit is used to determine the resource satisfaction information based on the first resource and the second resource.

9. The apparatus according to claim 8, wherein, The satisfaction unit includes: The completion subunit is used to determine the playback completion degree of the reference resource based on the current user's reference playback duration of the reference resource and the total duration of the reference resource. The contribution subunit is used to obtain the maximum playback duration of a single resource from the reference playback duration, and to determine the contribution of the reference resource based on the reference playback duration and the maximum playback duration. The satisfaction subunit is used to determine the satisfaction level based on the playback completion rate and the contribution rate.

10. The apparatus according to claim 9, wherein, The satisfaction sub-unit is specifically used to determine the satisfaction level using the following formula: ； in, For satisfaction, For playback completion rate, As for contribution level, It is a constant.

11. The apparatus according to any one of claims 8-10, wherein, The resource satisfaction unit includes: The resource vector subunit is used to average the vector representations of each of the first resources to obtain a first resource vector, and to average the vector representations of each of the second resources to obtain a second resource vector. The resource relevance subunit is used to determine a first relevance between the vector representation of the candidate resource and the first resource vector, and to determine a second relevance between the vector representation of the candidate resource and the second resource vector. The satisfaction coefficient subunit is used to subtract the second relevance from the first relevance to obtain the resource satisfaction coefficient at the current refresh depth; the resource satisfaction information is the resource satisfaction coefficient.

12. The apparatus according to claim 11, wherein, The recommended resource module includes: The new similarity submodule is used to multiply the resource satisfaction coefficient by the similarity to obtain a new similarity for the candidate resource; The resource filtering submodule is used to filter the candidate resources based on the new similarity.

13. The apparatus according to claim 8, wherein, The reference resource submodule includes: The resource quantity unit is used to determine the number of historical resources played by the current user at each of the historical refresh depths, based on the historical refresh depths of the historical resources. A refresh depth division unit is used to divide each of the historical refresh depths using the number of historical resources, to obtain at least two refresh depth intervals. A reference resource unit is used to determine the current refresh depth interval to which the current refresh depth belongs from the at least two refresh depth intervals; wherein, the reference resource is a historical resource whose historical refresh depth belongs to the current refresh depth interval.

14. The apparatus according to claim 13, wherein, The deviation between the number of historical resources in different refresh depth ranges is less than a preset deviation threshold.

15. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

16. 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 1-7.

17. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images