Recommendation methods, devices, equipment, and storage media for multimedia content

Abstract

This application relates to the field of computer technology and provides a method, apparatus, device, and storage medium for recommending multimedia content to solve the problem of low recommendation accuracy in recommendation systems. The method includes: receiving a target multimedia content request sent by an application server; obtaining a corresponding target historical ranking result set based on the account identifier carried in the target multimedia content request; then, coarsely ranking the first multimedia content set retrieved based on the target multimedia content request according to a first target ranking precision to obtain a second multimedia content set; and then, finely ranking the second multimedia content set and the target historical ranking result set according to a second target ranking precision to obtain a second ranking result; and selecting N target multimedia content items from the second ranking result and returning them to the application server. Fine ranking is performed only on the target historical ranking result set to avoid filtering out some historical multimedia content during the coarse ranking, thereby improving system accuracy.

Description

Technical Field

[0001] This application relates to the field of computer technology, and provides a method, apparatus, device, and storage medium for recommending multimedia content. Background Technology

[0002] In the era of big data, recommendation systems typically retrieve multimedia content from massive amounts of multimedia content that matches the multimedia content requests sent by the applications the user is using, and then sort the retrieved multimedia content to obtain a corresponding list of multimedia content.

[0003] However, if the above recommendation method is used, and the same application receives multiple multimedia requests for the same user within a short period of time, the probability of previously recommended multimedia content being recommended to that user again decreases. This will affect the accuracy of the recommendation system and reduce the precision of subsequent recommendations. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for recommending multimedia content to solve the problem of low recommendation accuracy in recommendation systems.

[0005] In a first aspect, embodiments of this application provide a method for recommending multimedia content, including:

[0006] Receive a target multimedia content request sent by the application server, and obtain a corresponding target historical sorting result set based on the account identifier carried in the target multimedia content request. Each target historical sorting result is generated by sorting multiple historical multimedia content according to a preset historical sorting precision based on a historical multimedia content request corresponding to the account identifier.

[0007] According to the first target sorting precision, a first sorting process is performed on the first multimedia content set retrieved based on the target multimedia content request, and a second multimedia content set is selected based on the first sorting result. According to the second target sorting precision, a second sorting process is performed on the second multimedia content set and the target historical sorting result set to obtain a second sorting result.

[0008] N target multimedia content items are selected from the second sorting result and returned to the application server, so that the application server sends the N target multimedia content items to the terminal, and the terminal displays the N target multimedia content items to the account corresponding to the account identifier. The sorting accuracy of the first target is lower than that of the second target.

[0009] Secondly, embodiments of this application also provide a multimedia content recommendation device, comprising:

[0010] The acquisition unit is used to receive a target multimedia content request sent by the application server, and to acquire a corresponding target historical sorting result set based on the account identifier carried by the target multimedia content request. Each target historical sorting result is generated by sorting multiple historical multimedia content according to a preset historical sorting precision based on a historical multimedia content request corresponding to the account identifier.

[0011] The sorting unit is configured to perform a first sorting process on a first set of multimedia content retrieved based on the target multimedia content request according to a first target sorting precision, and to filter out a second set of multimedia content based on the first sorting result, and to perform a second sorting process on the second set of multimedia content and the target historical sorting result set according to a second target sorting precision to obtain a second sorting result;

[0012] A transmission unit is configured to select N target multimedia content items from the second sorting result and return them to the application server, so that the application server sends the N target multimedia content items to the terminal, and the terminal displays the N target multimedia content items to the account corresponding to the account identifier, wherein the first target sorting accuracy is lower than the second target sorting accuracy.

[0013] Optionally, after obtaining the second sorting result, the device further includes a storage unit, the storage unit being used for:

[0014] Based on the second sorting result, the original historical sorting result set associated with the account identifier is updated to obtain a new original historical sorting result set.

[0015] Thirdly, embodiments of this application also provide a computer device, including a processor and a memory, wherein the memory stores program code that, when executed by the processor, causes the processor to perform the steps of any of the above-described recommended methods for multimedia content.

[0016] Fourthly, embodiments of this application also provide a computer-readable storage medium including program code, which, when the program product is run on a computer device, is used to cause the computer device to perform the steps of any of the above-described recommended methods for multimedia content.

[0017] The beneficial effects of this application are as follows:

[0018] This application provides a method, apparatus, device, and storage medium for recommending multimedia content. The method includes: receiving a target multimedia content request sent by an application server, and obtaining a corresponding target historical ranking result set based on the account identifier carried in the target multimedia content request, wherein each target historical ranking result is generated by ranking multiple historical multimedia content according to a preset historical ranking precision based on a historical multimedia content request corresponding to the account identifier; performing a first ranking process on the first multimedia content set recalled based on the target multimedia content request according to a first target ranking precision, and filtering out a second multimedia content set based on the first ranking result; and performing a second ranking process on the second multimedia content set and the target historical ranking result set according to a second target ranking precision to obtain a second ranking result; selecting N target multimedia content from the second ranking result and returning it to the application server, so that the application server sends the N target multimedia content to a terminal, and the terminal displays the N target multimedia content to the account corresponding to the account identifier, wherein the first target ranking precision is lower than the second target ranking precision. The multimedia content recommendation method provided in this application only performs fine sorting on the stored target historical ranking result set, preventing the target historical ranking result set from being filtered out in the coarse sorting process, thereby improving the accuracy and precision of the recommendation system.

[0019] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1a This is an optional schematic diagram of an application scenario in the embodiments of this application;

[0022] Figure 1b This is a schematic diagram of the architecture of a recommendation system;

[0023] Figure 2a A flowchart illustrating the recommended multimedia content provided in the embodiments of this application;

[0024] Figure 2b A flowchart illustrating the process of generating and storing the original historical sorting result set of account identifier i, as provided in the embodiments of this application;

[0025] Figure 2c A logical diagram illustrating the generation and storage of the original historical sorting result set of account identifier i provided in the embodiments of this application;

[0026] Figure 2d This is a logical diagram illustrating the acquisition of the target historical sorting result set for user A in this embodiment of the application.

[0027] Figure 2e This is a logical diagram illustrating the process of removing redundant historical multimedia content from the target historical sorting result set in this application embodiment.

[0028] Figure 3 This is a schematic diagram of the process by which the coarse sorting module performs the first sorting process on the first multimedia content set in an embodiment of this application;

[0029] Figure 4 This is a flowchart illustrating the process by which the fine-sorting module in this application performs a second sorting process on the second multimedia content set and the target historical sorting result set.

[0030] Figure 5a This is a schematic diagram of the process of recommending advertisements by the recommendation system in an embodiment of this application;

[0031] Figure 5b This is a logical diagram illustrating the recommendation system's ad recommendation process in an embodiment of this application.

[0032] Figure 6 A schematic diagram of the structure of a multimedia content recommendation device provided in an embodiment of this application;

[0033] Figure 7 This is a schematic diagram of the composition structure of a computer device provided in the embodiments of this application;

[0034] Figure 8 This is a schematic diagram of the structure of a computing device in an embodiment of this application. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0036] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein.

[0037] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0038] 1. Mixer: This refers to the central module of the recommendation system. Its main functions include, but are not limited to: obtaining user profiles of terminal users based on multimedia content requests sent by terminals, performing coarse and fine sorting on the recalled candidate multimedia content, supplementing the material information of multimedia content, and generating logs related to multimedia content requests.

[0039] 2. Predict Click-through Rate (pCTR): Calculate the account click probability for each candidate multimedia content in the candidate multimedia content set using the click-through rate prediction model in the recommendation system.

[0040] 3. Predict Conversion Rates (pCVR): Calculate the account conversion probability for each candidate multimedia content in the candidate multimedia content set using the conversion rate prediction model in the recommendation system.

[0041] 4. Gross Merchandise Volume (GMV): This refers to the amount of money paid for each piece of multimedia content delivered on a multimedia streaming platform. GMV is typically used to measure the overall traffic value of multimedia content.

[0042] 5. Expected revenue per thousand impressions (effective cost per mile, eCPM): This refers to the revenue generated from the exposure of multimedia content for every thousand plays.

[0043] 6. Recall: This refers to the recommendation system quickly retrieving a small subset of candidate multimedia content that an account may be interested in from a massive multimedia content library based on multimedia content requests.

[0044] 7. Coarse ranking: This refers to the recommendation system quickly filtering out multiple candidate multimedia content that meet the requirements from tens of thousands of recalled candidate multimedia content. In this embodiment, coarse ranking is described as the first ranking process.

[0045] 8. Fine ranking: This refers to the recommendation system's careful selection and ranking of multiple candidate multimedia content items to obtain the final N multimedia content items presented to the account. In this application embodiment, fine ranking is described as a second ranking process.

[0046] 9. Sorting precision: refers to the accuracy of the sorting. In the embodiments of this application, two sorting methods are included: coarse sorting with lower precision and fine sorting with higher precision.

[0047] The design concept of the embodiments of this application is briefly introduced below:

[0048] In the era of big data, recommendation systems typically retrieve multimedia content from massive amounts of multimedia content that matches the multimedia content requests sent by the applications the user is using, and then sort the retrieved multimedia content to obtain a corresponding list of multimedia content.

[0049] However, if the above recommendation method is used, and the same application receives multiple multimedia requests for the same user within a short period of time, the probability of previously recommended multimedia content being recommended to that user again decreases. This will affect the accuracy of the recommendation system and reduce the precision of subsequent recommendations.

[0050] In view of this, embodiments of this application provide a method for recommending multimedia content. The method includes: receiving a target multimedia content request sent by an application server; obtaining a corresponding target historical ranking result set based on the account identifier carried in the target multimedia content request, wherein each target historical ranking result is generated by ranking multiple historical multimedia content requests according to a preset historical ranking precision based on a historical multimedia content request corresponding to the account identifier; performing a first ranking process on the first multimedia content set recalled based on the target multimedia content request according to a first target ranking precision; filtering out a second multimedia content set based on the first ranking result; and performing a second ranking process on the second multimedia content set and the target historical ranking result set according to a second target ranking precision to obtain a second ranking result; selecting N target multimedia content items from the second ranking result and returning them to the application server, so that the application server sends the N target multimedia content items to a terminal, and the terminal displays the N target multimedia content items to the account corresponding to the account identifier, wherein the first target ranking precision is lower than the second target ranking precision.

[0051] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0052] like Figure 1aThe diagram illustrates an application scenario of an embodiment of this application. The scenario includes a physical terminal device 110, an application server 130, and a recommendation server 140 running a recommendation system. The recommendation system is used to recommend multimedia content to accounts. The physical terminal device 110 and the application server 130 can communicate via a communication network, as can the application server 130 and the recommendation server 140. In one optional implementation, the communication network can be a wired or wireless network. The physical terminal device 110 and the application server 130 can be directly or indirectly connected via wired or wireless communication, and similarly, the application server 130 and the recommendation server 140 can also be directly or indirectly connected via wired or wireless communication. This application does not impose any limitations on this connection.

[0053] When an account accesses the application server 130 through the access interface 120 of the physical terminal 110 (e.g., when a user browses web content through a browser installed on a mobile phone), the application server 130 sends a request for the target multimedia content to the recommendation server 140.

[0054] Recommendation server 140, based on the target multimedia content request, recalls the first multimedia content matching the request and, based on the account identifier carried in the target multimedia content request, obtains the corresponding target historical ranking result set. Recommendation server 140 first roughly sorts the recalled first multimedia content set according to the first target ranking precision, outputs the first ranking result, and filters the first ranking result to obtain a second multimedia content set; then, according to the second target ranking precision, it refines the second multimedia content set and the target historical ranking result, outputs the second ranking result, and selects N target multimedia content items from the second ranking result to return to application server 130.

[0055] The application server 130 then returns the N target multimedia contents to the physical terminal device 110, which then presents the N target multimedia contents on the access interface 120 to the account.

[0056] Specifically, in this application embodiment, the physical terminal device 110 is an electronic device used by the user. The electronic device can be a personal computer, mobile phone, tablet computer, laptop, e-book reader, smart home device, or other computer device with certain computing capabilities.

[0057] The application server 130 and recommendation server 140 in this embodiment can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. This application does not impose any limitations on these aspects. As disclosed in the multimedia content recommendation method of this application, multiple servers can form a blockchain, and the servers are nodes on the blockchain.

[0058] See Figure 1b The illustrated architecture diagram illustrates the components of the recommender system. Figure 1b As shown, the recommendation system comprises three parts: a mixer, a multimedia content retrieval module, and a coarse ranking module. The mixer itself consists of multiple modules, including a data acquisition module, a data storage module, and a fine ranking module. Specifically, the coarse ranking module includes a coarse ranking click-through rate prediction sub-model and a coarse ranking conversion rate prediction sub-model, while the fine ranking module includes a fine ranking click-through rate prediction sub-model and a fine ranking conversion rate prediction sub-model.

[0059] Specifically, mixer is the central module of the recommendation system. Its main functions include, but are not limited to: obtaining user profiles of terminal users based on multimedia content requests sent by the terminal, performing coarse sorting and fine sorting of the recalled multimedia content, supplementing the material information of the multimedia content, and generating logs related to multimedia content requests.

[0060] The data acquisition module is used to acquire target multimedia content requests sent by the application server, and based on the account identifier carried in the target multimedia content request, retrieve the corresponding target historical sorting result set from the data storage module. The multimedia content retrieval module is used to retrieve the first multimedia content that matches the target multimedia content request from a massive multimedia content library.

[0061] The coarse ranking module includes two sub-models: the coarse ranking click-through rate prediction sub-model and the coarse ranking conversion rate prediction sub-model. These sub-models extract features from each first multimedia content and its corresponding first set of related information. Based on the extracted features, they generate coarse ranking pCTR and coarse ranking pCVR for each first multimedia content. Then, based on the coarse ranking pCTR, coarse ranking pCVR, and GMV of each first multimedia content, they generate a first predicted recommendation value for each first multimedia content. According to the first predicted recommendation value of each first multimedia content, a first ranking result is generated. Finally, the first ranking result is filtered to obtain a second set of multimedia content.

[0062] The first associated information set of a first multimedia content x includes at least: the total number of exposures of the first multimedia content x, the positive exposure operation performed on the first multimedia content x by each historical account that has been pushed to the first multimedia content x, the number of times each historical account performed a positive exposure operation on the first multimedia content x, the negative exposure operation performed on the first multimedia content x by each historical account, and the number of times each historical account performed a negative exposure operation on the first multimedia content x.

[0063] A positive exposure operation refers to a historical account performing actions such as liking, commenting, sharing, or reporting on the first multimedia content x, while a negative exposure operation refers to a historical account only viewing the first multimedia content x without performing any actions on it.

[0064] The fine-ranking module's fine-ranking click-through rate prediction sub-model and fine-ranking conversion rate prediction sub-model extract features from each second multimedia content, each target historical multimedia content, and their corresponding second related information sets. Based on the extracted features, they generate fine-ranking pCTR and pCVR for each second multimedia content, as well as fine-ranking pCTR and pCVR for each target historical multimedia content. Then, based on the fine-ranking pCTR, pCVR, and GMV of each second multimedia content, and the fine-ranking pCTR, pCVR, and GMV of each target historical multimedia content, they generate second predicted recommendation values ​​for each second multimedia content and for each target historical multimedia content. According to the second predicted recommendation values ​​of each second multimedia content and the second predicted recommendation values ​​of each target historical multimedia content, a second ranking result is generated. Finally, the second ranking result is filtered to obtain N multimedia content.

[0065] The content of the second association information set is the same as that of the first association information set. Since the content of the first association information set has already been described above, the content of the second association information set will not be repeated here. Furthermore, the number of features extracted by the two sub-models in the coarse ranking module is less than the number of features extracted by the two sub-models in the fine ranking module. Therefore, the ranking accuracy of the coarse ranking module is lower than that of the fine ranking module.

[0066] See Figure 2a The flowchart shown illustrates the recommended method for multimedia content provided in the embodiments of this application.

[0067] S201: Receive the target multimedia content request sent by the application server, and obtain the corresponding target historical sorting result set based on the account identifier carried in the target multimedia content request. Each target historical sorting result is generated by sorting multiple historical multimedia content according to a preset historical sorting precision based on a historical multimedia content request corresponding to the account identifier.

[0068] To facilitate understanding, before introducing the specific implementation process of step 201, the process of generating a target historical ranking result will be described first. The multimedia content retrieval module retrieves the original historical multimedia content set based on a historical multimedia content request corresponding to an account identifier i; then, according to the first historical ranking precision of the coarse ranking module, the original historical multimedia content set is roughly ranked; and according to the second historical ranking precision of the fine ranking module, the candidate historical multimedia content set output by the coarse ranking module is finely ranked to obtain the target historical ranking result.

[0069] The generation and storage process of the original historical sorting result set for each account identifier is then described. (See also...) Figure 2b The flowchart shown Figure 2c The logic diagram shown illustrates the process of generating and storing the original historical sorting result set of account identifier i, using account identifier i as an example.

[0070] S2011: The data storage module obtains the second sorting result of account identifier i from the fine sorting module, and selects L candidate multimedia contents from the second sorting result according to the estimated storage amount of candidate multimedia contents.

[0071] The fine sorting module is used to select and sort multiple second multimedia content after the current round of coarse sorting, as well as multiple historical multimedia content output from previous rounds of fine sorting. Therefore, the L candidate multimedia content mentioned in step 2011 can be one or a combination of second multimedia content and historical multimedia content.

[0072] In this embodiment of the application, if the second sorting result is sorted from high to low according to the predicted recommendation value, the data storage module can select the top L candidate multimedia content from the second sorting result according to the estimated storage amount of the candidate multimedia content; if the second sorting result is sorted from low to high according to the predicted recommendation value, the data storage module can select the bottom L candidate multimedia content from the second sorting result according to the estimated storage amount of the candidate multimedia content.

[0073] S2012: The data storage module determines whether the original historical sorting result set of account identifier i is not stored. If yes, proceed to step 2013; otherwise, proceed to step 2014.

[0074] S2013: The data storage module establishes the original historical sorting result set for account identifier i, and stores L candidate multimedia contents as the first historical sorting result at the head of the original historical sorting result set.

[0075] S2014: The data storage module stores L candidate multimedia contents as the latest historical sorting results at the tail of the original historical sorting result set.

[0076] During step 2013 or step 2014, the data storage module also retrieves the generation times of L candidate multimedia content from the ranking module and stores the L candidate multimedia content along with their corresponding generation times in the data storage module. Typically, the generation times of the L candidate multimedia content in the same batch are the same.

[0077] S2015: If the actual storage amount of the historical sorting results in the new original historical sorting result set is greater than the estimated storage amount of the historical sorting results, then the data storage module will remove the historical sorting result at the head of the queue from the original historical sorting result set.

[0078] In this embodiment, the original historical sorting result set uses a "first-in, first-out" (FIFO) and "last-in, last-out" (LIFO) queue method to store and manage multiple historical sorting results for account identifier i. If the actual storage amount of historical sorting results in the new original historical sorting result set is greater than the estimated storage amount of historical sorting results, the historical sorting result at the head of the queue is determined to be expired data. The data storage module should remove the historical sorting result at the head of the queue from the original historical sorting result set to ensure data validity.

[0079] In addition, the same estimated storage capacity of candidate multimedia content and estimated storage capacity of historical ranking results can be set for each account in advance. Alternatively, the estimated storage capacity of candidate multimedia content and estimated storage capacity of historical ranking results can be set individually for each account according to the account's access activity level.

[0080] Next, we will introduce the process by which the data acquisition module obtains the target historical sorting result set.

[0081] The data acquisition module retrieves the original historical sorting result set corresponding to the account identifier carried in the target multimedia content request from the data storage module; the data acquisition module then retrieves the historical sorting result filtering conditions carried in the target multimedia content request, and outputs the original historical sorting result set that meets the historical sorting result filtering conditions as the target historical sorting result set.

[0082] The historical ranking results selection criteria include at least the following three types of information: time window information, estimated acquisition volume of historical ranking results, and estimated acquisition volume of historical multimedia content for each historical ranking result.

[0083] Specifically, the data acquisition module first filters out the original historical ranking results whose generation time matches the time window information from the original historical ranking result set according to the time window information in the historical ranking result filtering conditions, thus obtaining a candidate historical ranking result set; the data acquisition module then performs a second filtering on the candidate historical ranking result set according to the estimated acquisition volume of historical ranking results and the estimated acquisition volume of historical multimedia content for each historical ranking result in the historical ranking result filtering conditions, thus obtaining the target historical ranking result set.

[0084] In this embodiment of the application, it is necessary to ensure that the actual amount of historical ranking results obtained in the target historical ranking result set does not exceed the estimated amount of historical ranking results obtained, and to ensure that the actual amount of historical multimedia content obtained for each historical ranking result does not exceed the estimated amount of historical multimedia content obtained for each historical ranking result.

[0085] For example, see Figure 2d The illustrated logic diagram shows that User A's original historical ranking result set contains 5 historical ranking results, each containing 200 historical advertisements. The data acquisition module, based on the historical ranking result filtering criteria, retrieves historical ranking results 3-5 generated between 17:10 and 17:30 on May 5, 2021, from User A's original historical ranking result set. The number of historical multimedia content retrieved for each historical ranking result is 100. Finally, the 300 selected advertisements are output as User A's target historical ranking result set.

[0086] S202: According to the first target sorting precision, perform the first sorting process on the first multimedia content set recalled based on the target multimedia content request, and filter out the second multimedia content set based on the first sorting result. According to the second target sorting precision, perform the second sorting process on the second multimedia content set and the target historical sorting result set to obtain the second sorting result.

[0087] After step 201 is executed and before step 202 is executed, the data acquisition module and the multimedia content retrieval module are invoked to perform the following operations in parallel: the data acquisition module sends the obtained target historical sorting result set to the multimedia content retrieval module, and at the same time, the multimedia content retrieval module retrieves the first multimedia content set that matches the target multimedia content request from the massive multimedia content library.

[0088] The target multimedia content request includes, but is not limited to: account identifier, time window information, estimated acquisition volume of historical ranking results, estimated acquisition volume of historical multimedia content for each historical ranking result, application site identifier, application scene identifier, and application multimedia content display position identifier. The relationship between site, scene, and multimedia content display position is as follows: one application corresponds to one site, one site corresponds to multiple scenes (e.g., text / image scene, video scene), and each scene encompasses multiple multimedia content display positions.

[0089] The multimedia content retrieval module performs the following operations for each piece of historical multimedia content in the target historical ranking result set:

[0090] The multimedia content retrieval module matches a historical multimedia content with a first multimedia content set. If no matching item for the historical multimedia content is found in the first multimedia content set, the historical multimedia content is removed.

[0091] For example, see Figure 2e The logic diagram shown indicates that User A's target historical ranking result set contains 300 advertisements. If the 120th advertisement exists only in the target historical ranking result set and not in User A's first multimedia content set, then it is determined that the advertisement will stop being served in the application. Therefore, the advertisement should be removed from the target historical ranking result set.

[0092] When executing step 202, the coarse sorting module is first called to perform a rough sort, then the fine sorting module is called to perform a fine sort, and finally the fine sorting module outputs the second sorting result.

[0093] Specifically, see Figure 3 The flowchart shown illustrates that, according to the first target sorting precision of the coarse sorting module, the first sorting process is performed on the first multimedia content set.

[0094] S301: The coarse sorting module obtains the first association information set corresponding to each first multimedia content based on the first multimedia identifier of each first multimedia content in the first multimedia content set.

[0095] S302: The coarse ranking module then inputs each first multimedia content and its corresponding first set of related information into the preset first target ranking model to obtain the first predicted recommendation rate of each first multimedia content.

[0096] The first target ranking model includes a coarse-ranked click-through rate (CTR) prediction sub-model and a coarse-ranked conversion rate (CVR) prediction sub-model. The coarse-ranked CTR prediction sub-model and the coarse-ranked CVR prediction sub-model extract features from each of the first multimedia content pieces and their corresponding first set of related information. Based on the extracted features, they generate coarse-ranked pCTR and coarse-ranked pCVR for each of the first multimedia content pieces. The product of the coarse-ranked pCTR and the corresponding coarse-ranked pCVR is used as the first predicted recommendation rate for each of the first multimedia content pieces.

[0097] S303: The coarse ranking module obtains the first predicted recommendation value of each first multimedia content based on the first predicted recommendation rate and the corresponding value information.

[0098] The formula “eCPM1 = coarse ranking pCTR × coarse ranking pCVR × GMV” is used to determine the first predicted recommendation value for each first multimedia content.

[0099] S304: The coarse ranking module sorts each first multimedia content based on each first predicted recommendation value to obtain the first ranking result.

[0100] In this embodiment of the application, if the first sorting result is sorted from high to low according to the predicted recommendation value, then the first M multimedia content items in the first sorting result are used as the second multimedia content set; if the first sorting result is sorted from low to high according to the predicted recommendation value, then the first M multimedia content items in the first sorting result are used as the second multimedia content set.

[0101] Specifically, see Figure 4 The flowchart shown illustrates that, according to the second target sorting precision of the fine sorting module, the second sorting process is performed on the second multimedia content set and the target historical sorting result set.

[0102] S401: The fine-ranking module obtains the second association information set corresponding to each second multimedia content based on the second multimedia identifier of each second multimedia content in the second multimedia content set, and obtains the third association information set of each historical multimedia content based on the third multimedia identifier of each historical multimedia content in the target historical ranking result set.

[0103] S402: The fine ranking module inputs each second multimedia content and its corresponding second set of related information, each historical multimedia content and its corresponding third set of related information into a preset second target ranking model to obtain the second predicted recommendation rate of each second multimedia content and the second predicted recommendation rate of each historical multimedia content.

[0104] The second target ranking model includes a fine-grained click-through rate (CTR) prediction sub-model and a fine-grained conversion rate (CVR) prediction sub-model. These sub-models extract features from each second multimedia content, each target historical multimedia content, and their corresponding second related information sets. Based on these extracted features, they generate the fine-grained CTR and pCVR for each second multimedia content, as well as the fine-grained CTR and pCVR for each target historical multimedia content.

[0105] The product of the finely ranked pCTR of each second multimedia content and its corresponding finely ranked pCTR is used as the second predicted recommendation rate of each second multimedia content. The product of the finely ranked pCTR of each target historical multimedia content and its corresponding finely ranked pCVR is used as the second predicted recommendation rate of each target historical multimedia content.

[0106] S403: The fine ranking module obtains the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content based on the second predicted recommendation rate of each second multimedia content, the second predicted recommendation rate of each historical multimedia content, and the corresponding value information.

[0107] The formula “eCPM2 = fine-ranked pCTR × fine-ranked pCVR × GMV” is used to determine the second predicted recommendation value for each second multimedia content and the second predicted recommendation value for each historical multimedia content.

[0108] S404: The fine ranking module sorts each second multimedia content and each historical multimedia content based on the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content, and obtains the second ranking result.

[0109] Fine-ranking is performed only on each historical multimedia content within the target historical ranking result set to avoid some historical multimedia content being filtered out during the coarse-ranking process, thereby improving the accuracy and precision of the recommendation system. Furthermore, after the fine-ranking module obtains the second ranking result, the data storage module can update the original historical ranking result set associated with the account identifier based on the second ranking result, obtaining a new original historical ranking result set.

[0110] The second sorting result contains multimedia identifiers for each sequentially arranged multimedia content. In addition to storing the second sorting result, the data storage module also stores account identifiers, time window information, application site identifiers, application scene identifiers, and application multimedia content display position identifiers associated with the second sorting result.

[0111] S203: Select N target multimedia content items from the second sorting result and return them to the application server, so that the application server sends the N target multimedia content items to the terminal, and the terminal displays the N target multimedia content items to the account corresponding to the account identifier. The sorting accuracy of the first target is lower than that of the second target.

[0112] If the second sorting result is sorted from high to low according to the predicted recommendation value, the fine sorting module will output the top N multimedia content in the second sorting result as the target multimedia content; if the second sorting result is sorted from low to high according to the predicted recommendation value, the fine sorting module will output the bottom N multimedia content in the second sorting result as the target multimedia content.

[0113] When the multimedia content is an advertisement, see [link / reference]. Figure 5a The flowchart shown, and Figure 5b The illustrated logic diagram, using a specific embodiment, introduces a method for recommending advertisements in a recommendation system.

[0114] S501: The access layer of the recommendation system transforms the target ad request sent by the application server into a new target ad request that conforms to the internal processing protocol, and sends the new target ad request to the mixer.

[0115] S502: The data acquisition module in mixer retrieves the corresponding set of original historical sorting results from the data storage module based on the account identifier carried in the new target ad content request;

[0116] S503: The data acquisition module obtains a set of candidate historical sorting results from the original historical sorting results based on the time window information in the filtering conditions of the historical sorting results.

[0117] S504: The data acquisition module then obtains the target historical ranking result set from the candidate historical ranking result set based on the estimated acquisition volume of historical ranking results in the historical ranking result filtering conditions and the estimated acquisition volume of historical advertisements for each historical ranking result.

[0118] S505: The data acquisition module sends the obtained target historical ranking result set to the multimedia content recall module. At the same time, the multimedia content recall module recalls the first set of advertisements that match the new target advertisement request from the massive advertisement library.

[0119] S506: The multimedia content retrieval module filters the target historical ranking result set to obtain a new target historical ranking result set;

[0120] S507: The multimedia content recall module sends the first set of advertisements to the coarse ranking module and sends the new target historical ranking result set to the fine ranking module;

[0121] S508: After the coarse sorting module performs a coarse sort on the first set of advertisements, it obtains the second set of advertisements and sends the second set of advertisements to the fine sorting module;

[0122] S509: The fine sorting module performs a fine sorting on the second set of advertisements and the new target historical sorting result set, obtains the second sorting result, and sends the second sorting result to the data storage module for storage.

[0123] S510: The fine sorting module selects N target advertisements from the second sorting results and returns them to the application server, so that the application server sends the N target advertisements to the terminal, and the terminal displays the N target advertisements to the account corresponding to the account identifier in the corresponding advertisement display position.

[0124] Based on the same inventive concept as the above-described method embodiments, this application also provides a schematic diagram of the structure of a multimedia content recommendation device. For example... Figure 6 As shown, the multimedia content recommendation device 600 may include:

[0125] The acquisition unit 601 is used to receive a target multimedia content request sent by the application server, and to acquire a corresponding target historical sorting result set based on the account identifier carried in the target multimedia content request. Each target historical sorting result is generated by sorting multiple historical multimedia content according to a preset historical sorting precision based on a historical multimedia content request corresponding to the account identifier.

[0126] The sorting unit 602 is used to perform a first sorting process on a first multimedia content set retrieved based on a target multimedia content request according to a first target sorting precision, and to filter out a second multimedia content set based on the first sorting result, and to perform a second sorting process on the second multimedia content set and the target historical sorting result set according to a second target sorting precision to obtain a second sorting result;

[0127] The transmission unit 603 is used to select N target multimedia content from the second sorting result and return them to the application server, so that the application server sends the N target multimedia content to the terminal, and the terminal displays the N target multimedia content to the account corresponding to the account identifier. The sorting accuracy of the first target is lower than that of the second target.

[0128] Optionally, the acquisition unit 601 is used for:

[0129] Based on the account identifier carried in the target multimedia content request, obtain the original historical sorting result set corresponding to the account identifier;

[0130] Obtain the historical sorting result filtering conditions carried by the target multimedia content request, and output the original historical sorting result set that meets the historical sorting result filtering conditions as the target historical sorting result set.

[0131] Optionally, the acquisition unit 601 is used for:

[0132] Based on the time window information in the historical sorting result filtering conditions, the original historical sorting results whose generation time matches the time window information are filtered from the original historical sorting result set to obtain the candidate historical sorting result set.

[0133] Based on the estimated acquisition volume of historical ranking results and the estimated acquisition volume of historical multimedia content for each historical ranking result in the filtering criteria of historical ranking results, a secondary filtering is performed on the candidate historical ranking result set to obtain the target historical ranking result set.

[0134] Optionally, after obtaining the corresponding target historical sorting result set, and before performing the first sorting process on the first multimedia content set requested for recall according to the first target sorting precision, the obtaining unit 601 is further configured to:

[0135] For each historical multimedia content in the target historical ranking result set, perform the following operations: match a historical multimedia content with the first multimedia content set; if no matching item for a historical multimedia content is found in the first multimedia content set, remove a historical multimedia content.

[0136] Optionally, sorting unit 602 is used for:

[0137] Based on the first multimedia identifier of each first multimedia content in the first multimedia content set, obtain the first association information set corresponding to each first multimedia content;

[0138] Each first multimedia content and its corresponding first set of related information are input into a preset first target ranking model to obtain the first predicted recommendation rate of each first multimedia content.

[0139] Based on the first predicted recommendation rate of each first multimedia content and the corresponding value information, the first predicted recommendation value of each first multimedia content is obtained.

[0140] Based on each first predicted recommendation value, the first multimedia content is sorted to obtain the first ranking result.

[0141] Optionally, sorting unit 602 is used for:

[0142] Based on the second multimedia identifier of each second multimedia content in the second multimedia content set, obtain the second association information set corresponding to each second multimedia content; and based on the third multimedia identifier of each historical multimedia content in the target historical sorting result set, obtain the third association information set of each historical multimedia content.

[0143] Each second multimedia content and its corresponding second set of associated information, each historical multimedia content and its corresponding third set of associated information are input into a preset second target ranking model to obtain the second predicted recommendation rate of each second multimedia content and the second predicted recommendation rate of each historical multimedia content.

[0144] Based on the second predicted recommendation rate of each second multimedia content, the second predicted recommendation rate of each historical multimedia content, and the corresponding value information, the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content are obtained.

[0145] Based on the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content, the second multimedia content and each historical multimedia content are sorted to obtain the second ranking result.

[0146] Optionally, after obtaining the second sorting result, the device 600 further includes a storage unit 604, which is used for:

[0147] Based on the second sorting result, the original historical sorting result set associated with the account identifier is updated to obtain a new original historical sorting result set.

[0148] For ease of description, the above sections are divided into modules (or units) according to their functions and described separately. Of course, in implementing this application, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0149] Having introduced the recommended methods and apparatus for multimedia content according to exemplary embodiments of this application, we will now introduce a computer device according to another exemplary embodiment of this application.

[0150] Those skilled in the art will understand that various aspects of this application can be implemented as a system, method, or program product. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as a "circuit," "module," or "system."

[0151] Based on the same inventive concept as the above-described method embodiments, this application also provides a computer device, see below. Figure 7 As shown, the computer device 700 may include at least a processor 701 and a memory 702. The memory 702 stores program code that, when executed by the processor 701, causes the processor 701 to perform the steps of any of the recommended methods for multimedia content described above.

[0152] In some possible implementations, the computing device according to this application may include at least one processor and at least one memory. The memory stores program code that, when executed by the processor, causes the processor to perform the steps in the recommended methods for multimedia content according to various exemplary embodiments of this application described above. For example, the processor may perform actions such as... Figure 2a The steps are shown in the figure.

[0153] The following reference Figure 8 To describe a computing device 800 according to this embodiment of the present application. Figure 8 The computing device 800 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0154] like Figure 8 As shown, the computing device 800 is presented in the form of a general-purpose computing device. The components of the computing device 800 may include, but are not limited to: at least one processing unit 801, at least one storage unit 802, and a bus 803 connecting different system components (including storage unit 802 and processing unit 801).

[0155] Bus 803 represents one or more of several bus structures, including a memory bus or memory controller, peripheral bus, processor, or a local bus using any of the various bus structures.

[0156] Storage unit 802 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 8021 and / or cache memory unit 8022, and may further include read-only memory (ROM) 8023.

[0157] Storage unit 802 may also include a program / utility 8025 having a set (at least one) program module 8024, such program module 8024 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0158] The computing device 800 can also communicate with one or more external devices 804 (e.g., keyboard, pointing device, etc.), and with one or more devices that enable a user to interact with the computing device 800, and / or with any device that enables the computing device 800 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via the input / output (I / O) interface 805. Furthermore, the computing device 800 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via a network adapter 806. Figure 8 As shown, network adapter 806 communicates with other modules for computing device 800 via bus 803. It should be understood that, although... Figure 8 As not shown, the computing device 800 may be used with other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0159] Based on the same inventive concept as the above-described method embodiments, various aspects of the multimedia content recommendation method provided in this application can also be implemented as a program product, which includes program code. When the program product is run on a computer device, the program code is used to cause the computer device to perform the steps in the multimedia content recommendation method according to the various exemplary embodiments of this application described above. For example, the computer device can perform actions such as... Figure 2a The steps are shown in the figure.

[0160] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable 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 thereof.

[0161] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0162] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for recommending multimedia content, characterized in that, include: Receive a target multimedia content request sent by the application server, and obtain a corresponding target historical sorting result set based on the account identifier carried in the target multimedia content request. Each target historical sorting result is generated by sorting multiple historical multimedia content according to a preset historical sorting precision based on a historical multimedia content request corresponding to the account identifier. According to the first target sorting precision, the first set of multimedia content retrieved based on the target multimedia content request is sorted out, and the second set of multimedia content is selected based on the first sorting result. Based on the second multimedia identifier of each second multimedia content in the second multimedia content set, obtain the second association information set corresponding to each second multimedia content; and based on the third multimedia identifier of each historical multimedia content in the target historical sorting result set, obtain the third association information set of each historical multimedia content. Each second multimedia content and its corresponding second set of associated information, each historical multimedia content and its corresponding third set of associated information are input into a preset second target ranking model to obtain the second predicted recommendation rate of each second multimedia content and the second predicted recommendation rate of each historical multimedia content. Based on the second predicted recommendation rate of each second multimedia content, the second predicted recommendation rate of each historical multimedia content, and the corresponding value information, the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content are obtained. Based on the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content, the second multimedia content and the historical multimedia content are sorted to obtain a second sorting result. N target multimedia content items are selected from the second sorting result and returned to the application server, so that the application server sends the N target multimedia content items to the terminal, and the terminal displays the N target multimedia content items to the account corresponding to the account identifier. The sorting accuracy of the first target is lower than that of the second target.

2. The method as described in claim 1, characterized in that, The step of obtaining the corresponding target historical ranking result set based on the account identifier carried in the target multimedia content request includes: Based on the account identifier carried in the target multimedia content request, obtain the original historical sorting result set corresponding to the account identifier; Obtain the historical sorting result filtering conditions carried by the target multimedia content request, and output the original historical sorting result set that meets the historical sorting result filtering conditions as the target historical sorting result set.

3. The method as described in claim 2, characterized in that, The step of outputting the set of original historical sorting results that meets the historical sorting result filtering conditions as the target historical sorting result set includes: According to the time window information in the historical sorting result filtering conditions, the original historical sorting results whose generation time conforms to the time window information are filtered from the original historical sorting result set to obtain the candidate historical sorting result set. The candidate historical ranking result set is further filtered according to the estimated acquisition volume of historical ranking results and the estimated acquisition volume of historical multimedia content for each historical ranking result in the filtering conditions of the historical ranking results, so as to obtain the target historical ranking result set.

4. The method as described in claim 1, characterized in that, After obtaining the corresponding target historical ranking result set, and before performing the first ranking process on the first multimedia content set requested for recall according to the first target ranking precision, the method further includes: For each historical multimedia content in the target historical sorting result set, the following operations are performed respectively: match a historical multimedia content with the first multimedia content set; if no matching item for the historical multimedia content is found in the first multimedia content set, then remove the historical multimedia content.

5. The method as described in claim 1, characterized in that, The first sorting process, performed on the first multimedia content set according to the first target sorting precision, includes: Based on the first multimedia identifier of each first multimedia content in the first multimedia content set, obtain the first association information set corresponding to each first multimedia content; Each of the first multimedia content and its corresponding first association information set are input into a preset first target ranking model to obtain the first predicted recommendation rate of each of the first multimedia content. Based on the first predicted recommendation rate of each first multimedia content and the corresponding value information, the first predicted recommendation value of each first multimedia content is obtained. Based on each first predicted recommendation value, the first multimedia content is sorted to obtain the first sorting result.

6. The method according to any one of claims 1 to 5, characterized in that, After obtaining the second sorting result, the following is also included: Based on the second sorting result, the original historical sorting result set associated with the account identifier is updated to obtain a new original historical sorting result set.

7. A multimedia content recommendation device, characterized in that, include: The acquisition unit is used to receive a target multimedia content request sent by the application server, and to acquire a corresponding target historical sorting result set based on the account identifier carried by the target multimedia content request. Each target historical sorting result is generated by sorting multiple historical multimedia content according to a preset historical sorting precision based on a historical multimedia content request corresponding to the account identifier. The sorting unit is used to perform a first sorting process on a first set of multimedia content retrieved based on the target multimedia content request according to a first target sorting precision, and to filter out a second set of multimedia content based on the first sorting result. Based on the second multimedia identifier of each second multimedia content in the second multimedia content set, obtain the second association information set corresponding to each second multimedia content; and based on the third multimedia identifier of each historical multimedia content in the target historical sorting result set, obtain the third association information set of each historical multimedia content. Each second multimedia content and its corresponding second set of associated information, each historical multimedia content and its corresponding third set of associated information are input into a preset second target ranking model to obtain the second predicted recommendation rate of each second multimedia content and the second predicted recommendation rate of each historical multimedia content. Based on the second predicted recommendation rate of each second multimedia content, the second predicted recommendation rate of each historical multimedia content, and the corresponding value information, the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content are obtained. Based on the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content, the second multimedia content and the historical multimedia content are sorted to obtain a second sorting result. A transmission unit is configured to select N target multimedia content items from the second sorting result and return them to the application server, so that the application server sends the N target multimedia content items to the terminal, and the terminal displays the N target multimedia content items to the account corresponding to the account identifier, wherein the first target sorting accuracy is lower than the second target sorting accuracy.

8. The apparatus as claimed in claim 7, characterized in that, The acquisition unit is used for: Based on the account identifier carried in the target multimedia content request, obtain the original historical sorting result set corresponding to the account identifier; Obtain the historical sorting result filtering conditions carried by the target multimedia content request, and output the original historical sorting result set that meets the historical sorting result filtering conditions as the target historical sorting result set.

9. The apparatus as claimed in claim 8, characterized in that, The acquisition unit is used for: According to the time window information in the historical sorting result filtering conditions, the original historical sorting results whose generation time conforms to the time window information are filtered from the original historical sorting result set to obtain the candidate historical sorting result set. The candidate historical ranking result set is further filtered according to the estimated acquisition volume of historical ranking results and the estimated acquisition volume of historical multimedia content for each historical ranking result in the filtering conditions of the historical ranking results, so as to obtain the target historical ranking result set.

10. The apparatus as claimed in claim 7, characterized in that, After obtaining the corresponding target historical sorting result set, and before performing the first sorting process on the first multimedia content set requested for recall according to the first target sorting precision, the obtaining unit is further configured to: For each historical multimedia content in the target historical sorting result set, the following operations are performed respectively: match a historical multimedia content with the first multimedia content set; if no matching item for the historical multimedia content is found in the first multimedia content set, then remove the historical multimedia content.

11. The apparatus as claimed in claim 7, characterized in that, The sorting unit is used for: Based on the second multimedia identifier of each second multimedia content in the second multimedia content set, obtain the second association information set corresponding to each second multimedia content; and based on the third multimedia identifier of each historical multimedia content in the target historical sorting result set, obtain the third association information set of each historical multimedia content. Each second multimedia content and its corresponding second set of associated information, each historical multimedia content and its corresponding third set of associated information are input into a preset second target ranking model to obtain the second predicted recommendation rate of each second multimedia content and the second predicted recommendation rate of each historical multimedia content. Based on the second predicted recommendation rate of each second multimedia content, the second predicted recommendation rate of each historical multimedia content, and the corresponding value information, the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content are obtained. Based on the second predicted recommendation value of each second multimedia content and the second predicted recommendation value of each historical multimedia content, the second multimedia content and the historical multimedia content are sorted to obtain the second sorting result.

12. A computer device, characterized in that, It includes a processor and a memory, wherein the memory stores program code that, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 6.

13. A computer-readable storage medium, characterized in that, It includes program code that, when the program product is run on a computer device, causes the computer device to perform the steps of the method according to any one of claims 1 to 6.

Images