Multimedia content recommendation method, apparatus, device, and storage medium
By detecting user browsing behavior and estimating retention rates, and optimizing multimedia content recommendations, the problem of insufficient user retention in existing systems has been solved, resulting in effective improvement in retention rates and user experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DOUYIN VISION CO LTD
- Filing Date
- 2020-07-27
- Publication Date
- 2026-06-19
AI Technical Summary
Existing recommendation systems have failed to effectively improve user retention rates, resulting in insufficient daily active users.
By detecting user browsing behavior and estimating retention rates based on historical access records, the estimated retention rate is weighted with other recommendation metrics to determine the multimedia content to be recommended.
This improved user retention, which indirectly increased click-through rate and conversion rate, thus enhancing the user experience.
Smart Images
Figure CN113987222B_ABST
Abstract
Description
Technical Field
[0001] This application relates to information processing technology, and in particular to a method, apparatus, device and storage medium for recommending multimedia content. Background Technology
[0002] Recommendation systems often use click-through rate (CTR) or conversion rate (CVR) to recommend content and improve those metrics. However, current recommendation methods fail to improve user retention, while the core objective of recommendation systems is to increase daily active users through user retention—a problem that current systems haven't addressed. Summary of the Invention
[0003] To address the aforementioned issues, this invention proposes a multimedia content recommendation method, apparatus, device, and storage medium. This enables multimedia content recommendation based on estimated retention rates, laying the foundation for improving retention rates.
[0004] In a first aspect, embodiments of this application provide a multimedia content recommendation method, including:
[0005] The user's browsing behavior on the first multimedia content in the application was detected;
[0006] Based on the historical browsing behavior of the multimedia content accessed as indicated by the historical access records, the estimated retention rate of the user when accessing the application again is obtained; wherein, the estimated retention rate is associated with the content characteristics of the first multimedia content.
[0007] Based at least on the estimated retention rate, determine the multimedia content to be recommended to the user.
[0008] In a specific example of this application, obtaining the estimated retention rate of the user's subsequent access to the application based on the historical browsing behavior of accessed multimedia content indicated by historical access records includes:
[0009] Based on the historical access records, the user's historical browsing behavior of accessed multimedia content at a historical time and the retention behavior at the next time after the historical time are obtained, and retention value features that characterize the contribution of the content features of the accessed multimedia content to the retention behavior are determined.
[0010] Based on the retention value characteristics and the content characteristics of the first multimedia content, the estimated retention rate of the user who visits the application again after browsing the first multimedia content is determined.
[0011] In a specific example of this application, the determination of retention value features, which characterize the contribution of content features of the accessed multimedia content to retention behavior, includes:
[0012] Determine the mapping relationship between the user's historical browsing behavior and the retention behavior at the next time point in the historical time, input the mapping relationship into the prediction model, and obtain the retention value feature, which characterizes the contribution of the content features of the accessed multimedia content to the retention behavior.
[0013] In a specific example of this application, determining the multimedia content to be recommended to the user based at least on the estimated retention rate includes:
[0014] Obtain the estimated recommendation metrics corresponding to the user, wherein the preset recommendation metrics are other recommendation metrics other than the estimated retention rate;
[0015] The estimated retention rate is weighted with the obtained estimated recommendation index, and the multimedia content to be recommended to the user is determined based on the recommendation index obtained after weighting.
[0016] In a specific example of this application, determining the multimedia content to be recommended to the user based at least on the estimated retention rate includes:
[0017] When there are multiple first multimedia content items, the estimated retention rates corresponding to the multiple first multimedia content items are sorted, and the multimedia content to be recommended to the user is determined based on the sorting results.
[0018] In a specific example of this application, the method further includes:
[0019] Once the application is detected to have been accessed by the user again, multimedia content intended for that user will be recommended to that user.
[0020] Secondly, embodiments of this application provide a multimedia content recommendation device, including:
[0021] The detection unit is used to detect the user's browsing behavior on the first multimedia content in the application;
[0022] The estimation unit is used to obtain the estimated retention rate of the user's re-access to the application based on the historical browsing behavior of the accessed multimedia content indicated by the historical access records; wherein the estimated retention rate is associated with the content characteristics of the first multimedia content.
[0023] The recommendation unit is used to determine multimedia content to be recommended to the user, based at least on the estimated retention rate.
[0024] In a specific example of this application, the estimation unit is further configured to:
[0025] Based on the historical access records, the user's historical browsing behavior of accessed multimedia content at a historical time and the retention behavior at the next time after the historical time are obtained, and retention value features that characterize the contribution of the content features of the accessed multimedia content to the retention behavior are determined.
[0026] Based on the retention value characteristics and the content characteristics of the first multimedia content, the estimated retention rate of the user who visits the application again after browsing the first multimedia content is determined.
[0027] In a specific example of this application, the estimation unit is further configured to:
[0028] Determine the mapping relationship between the user's historical browsing behavior and the retention behavior at the next time point in the historical time, input the mapping relationship into the prediction model, and obtain the retention value feature, which characterizes the contribution of the content features of the accessed multimedia content to the retention behavior.
[0029] In a specific example of this application, the recommendation unit is further configured to:
[0030] Obtain the estimated recommendation metrics corresponding to the user, wherein the preset recommendation metrics are other recommendation metrics other than the estimated retention rate;
[0031] The estimated retention rate is weighted with the obtained estimated recommendation index, and the multimedia content to be recommended to the user is determined based on the recommendation index obtained after weighting.
[0032] In a specific example of this application, the recommendation unit is further configured to:
[0033] When there are multiple first multimedia content items, the estimated retention rates corresponding to the multiple first multimedia content items are sorted, and the multimedia content to be recommended to the user is determined based on the sorting results.
[0034] In a specific example of this application, the recommendation unit is further configured to:
[0035] Once the application is detected to have been accessed by the user again, multimedia content intended for that user will be recommended to that user.
[0036] Thirdly, embodiments of this application provide a multimedia content recommendation device, including:
[0037] One or more processors;
[0038] A memory that is communicatively connected to the one or more processors;
[0039] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to perform the methods described above.
[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.
[0041] This provides a complete scheme for determining the estimated retention rate, which targets the probability of re-accessing the application after accessing the first multimedia content. Therefore, it lays the foundation for effectively improving retention rates. Simultaneously, it enables recommendation based on retention rates. Attached Figure Description
[0042] Figure 1 This is a schematic diagram illustrating the implementation process of the multimedia content recommendation method according to an embodiment of the present invention;
[0043] Figure 2 This is a schematic diagram of the retention curve in the embodiment of this application;
[0044] Figure 3 This is a schematic diagram illustrating an application scenario of an embodiment of this application;
[0045] Figure 4 This is a schematic diagram illustrating the implementation process of the multimedia content recommendation method of this invention in a specific application scenario.
[0046] Figure 5 This is a schematic diagram of the structure of the multimedia content recommendation device according to an embodiment of the present invention;
[0047] Figure 6 A schematic diagram of the structure of a multimedia content recommendation device according to an embodiment of the present invention. Detailed Implementation
[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0049] In some of the processes described in the specification, claims and accompanying drawings of this application, there are multiple operations that occur in a specific order. However, it should be clearly understood that these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.
[0050] In industrial systems (such as recommendation systems), retention rate is a core product metric, signifying its attractiveness to users. Most product iterations and improvements focus on optimizing this metric. However, due to the complexity of industrial systems and limitations in technological development, there are currently no precedents for directly modeling and optimizing the retention metric (i.e., retention rate). Instead, modeling is typically done for certain key scenarios within the recommendation system, such as click-through rate and conversion rate in feed (continuously updated content presented to users) recommendations. Improving click-through rate and conversion rate indirectly improves retention rate. However, simply improving click-through rate and conversion rate leads to insufficient information mining. Furthermore, since retention rate and click-through rate / conversion rate are not entirely correlated, merely improving click-through rate and conversion rate does not effectively improve retention rate.
[0051] Based on this, embodiments of this application provide a multimedia content recommendation method, apparatus, device, and storage medium; specifically, Figure 1 This is a schematic diagram illustrating the implementation process of the multimedia content recommendation method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:
[0052] Step 101: The multimedia content recommendation device detects the user's browsing behavior of the first multimedia content in the application.
[0053] Step 102: The multimedia content recommendation device obtains the estimated retention rate of the user's re-access to the application based on the historical browsing behavior of the multimedia content accessed as indicated by the historical access records; wherein, the estimated retention rate is associated with the content characteristics of the first multimedia content.
[0054] Step 103: The multimedia content recommendation device determines the multimedia content to be recommended to the user, based at least on the estimated retention rate.
[0055] like Figure 2 As shown, the trend of retention rate, i.e., the retention curve (as shown) Figure 2 The left graph in the image typically follows a power function; any retention rate trend is a power function f(x) = ax. b , (x=1,2,3...), where a and b are constants.
[0056] In practical applications, the determined estimated retention rate, associated with the first multimedia content, can specifically include: the determined estimated retention rate being associated with the content features of the first multimedia content. Thus, after determining the estimated retention rate associated with the first multimedia content, the multimedia content recommendation device determines whether to continue recommending multimedia content associated with the content features of the first multimedia content to the user, thereby improving user retention within the application.
[0057] This solution, by utilizing the historical browsing behavior of previously accessed multimedia content indicated by historical access records after detecting the first multimedia content viewed in the application, can obtain the estimated retention rate of the user's subsequent visits to the application. In other words, it predicts the user's expected retention rate upon revisiting the application. Thus, a complete scheme for determining the estimated retention rate is provided, and this estimated retention rate targets the probability of revisiting the application after accessing the first multimedia content. Therefore, it lays the foundation for effectively improving retention rates. Simultaneously, it enables recommendation based on retention rates.
[0058] In practical applications, the first multimedia content can specifically be multimedia content recommended by the application to the user. In this case, the solution of this application can predict the probability that the user will revisit the application after browsing the first multimedia content recommended by the application (i.e., the estimated retention rate). This lays the foundation for effectively and directly improving the retention rate. Furthermore, in practical scenarios, the first multimedia content can not only be the multimedia content recommended by this application based on the retention rate, but also the multimedia content recommended based on other recommendation metrics, such as click-through rate or conversion rate.
[0059] In a specific example of this application, the estimated retention rate can be obtained in the following manner: based on the historical access records, the user's historical browsing behavior of accessed multimedia content at a historical time and the retention behavior at the next time after the historical time are obtained, and a retention value feature is determined to characterize the contribution of the content features of the accessed multimedia content to the retention behavior; based on the retention value feature and the content features of the first multimedia content, the estimated retention rate of the user accessing the application again after browsing the first multimedia content is determined.
[0060] Here, retention behavior can specifically include: after a user accessed specific multimedia content in the application at a historical time, the user accessed multimedia content in the application again at a later time; or, after a user accessed specific multimedia content in the application at a historical time, the user did not access multimedia content in the application at a later time.
[0061] In practical applications, the historical time can be any time recorded for the user in the historical access records, such as any natural day. In this case, the next time can be the day after that natural day. Of course, in practical applications, the method of determining the next time of the historical time is related to the method of calculating the retention rate, and can be set according to the actual statistical method. This application does not impose any restrictions on this.
[0062] Thus, since the retention rate can be estimated based on the retention value feature, and the retention value feature can characterize the contribution of the content features of the accessed multimedia content to the retention behavior, a foundation is laid for effectively improving the retention rate.
[0063] In a specific example of this application, a prediction model can also be used to obtain retention value features, specifically including: determining the mapping relationship between the user's historical browsing behavior and the retention behavior at the next time after the historical time, inputting the mapping relationship into the prediction model, and obtaining retention value features that characterize the contribution of the content features of the accessed multimedia content to the retention behavior.
[0064] In this way, by using a pre-trained prediction model to obtain retention value characteristics, the accuracy of the predicted retention rate is improved, laying the foundation for effectively improving the retention rate.
[0065] In one example, the mapping relationship can be specifically a mapping relationship between the content characteristics of the accessed multimedia content corresponding to the historical browsing behavior and the retention behavior.
[0066] In a specific example of this application, when determining recommended content, not only the estimated retention rate metric can be referenced, but other recommendation metrics can also be considered. For example, the estimated recommendation metrics corresponding to the user can be obtained. Here, the preset recommendation metrics are other recommendation metrics besides the estimated retention rate. Then, the estimated retention rate and the obtained estimated recommendation metrics are weighted, and the multimedia content to be recommended to the user is determined based on the weighted recommendation metrics. In this way, by flexibly configuring the weights of different estimated recommendation metrics, the application scenario can be flexibly adapted, enhancing the overall value of the solution in this application and laying the foundation for improving user experience.
[0067] In a specific example of this application, when there are multiple first multimedia content, multiple estimated retention rates can be obtained. The multiple estimated retention rates are sorted, and the multimedia content to be recommended to the user is determined based on the sorting results, for example, sorted in descending order. A preset number of recommended content at the top of the sorting are then recommended to the user as the final recommended content, thereby improving the retention rate.
[0068] In a specific example of this application, recommendations can be made in two ways: multimedia content to be recommended to the user can be directly recommended to the user, for example, directly recommended to the terminal used by the user; or, after detecting that the user has accessed the application again, multimedia content to be recommended to the user can be recommended to the user. Since the multimedia content recommended to the user is based on the estimated retention rate, and the estimated retention rate is based on historical browsing behavior, it can effectively improve the retention rate and enhance the user experience.
[0069] This provides a complete scheme for determining the estimated retention rate, which targets the probability of re-accessing the application after accessing the first multimedia content. Therefore, it lays the foundation for effectively improving retention rates. Simultaneously, it enables recommendation based on retention rates.
[0070] The following detailed example illustrates this solution in further detail. Specifically, this example applies the solution to a recommendation or search system. Based on users' historical behavioral data (i.e., historical access records), such as access and search behavior data on a specific day, the system predicts customer retention rates over a future set time period. This helps determine which recommendation behaviors are beneficial and detrimental to user retention. Finally, the retention rates are ranked, and recommendations are made based on the ranking results, effectively improving user retention and enhancing the user experience within the system. Furthermore, the increased retention rate indirectly improves click-through rate and conversion rate.
[0071] In practical scenarios, the solution proposed in this application is applied to clusters of recommendation systems or search systems. For example, Figure 3 As shown, the proposed solution runs on the server-side cluster side, and after the server determines the recommended multimedia content, it sends the multimedia content to the terminal side (such as laptops, Kindles, mobile phones, etc.).
[0072] The retention rate in this example is the next-day retention rate; correspondingly, the next-day retention rate can be estimated. Of course, in practical applications, different systems may have different definitions of retention rate. For example, it may be necessary to estimate the retention rate on day nm. This application does not impose any restrictions on this and only uses the next-day retention rate as an example for explanation.
[0073] Specifically, such as Figure 4As shown, the main steps of this example include: First, the data acquisition unit collects data based on the user's historical behavior data to obtain the correspondence between the user's first-day access behavior and second-day retention behavior, i.e., obtaining sample data. Then, the sample data is transmitted to the model prediction unit through the data stream unit. Here, the model prediction unit can be an online prediction unit, for example, using discrete features for online prediction, or using a deep learning model for online prediction. The model prediction unit uses the model prediction unit to obtain the retention prediction score (i.e., the predicted retention rate) corresponding to the multimedia content currently accessed by the user. This can characterize the value of the currently accessed multimedia content category to retention behavior (i.e., retention value characteristics), thereby obtaining the estimated retention rate. Of course, in real-world scenarios, the system can also integrate and calculate the estimated retention rate with other estimated recommendation metrics from scenarios such as clicks and conversions. For example, weighting each metric, where the weight of each metric can be set based on the specific scenario. After integration and calculation, the comprehensive estimated recommendation metrics are sorted, and all sorted recommendation categories are fed back to the terminal, or the multimedia content corresponding to all sorted recommendation categories is fed back to the terminal. Alternatively, only categories that meet specific conditions after sorting can be used as recommendation types; this application does not impose such restrictions. Accordingly, the above steps can be implemented through the following five units, specifically...
[0074] The data collection unit needs to be able to determine user retention on the third day, as data retention for the following day is only known on the third day. This is because retained data is often delayed. For example, if a user viewed a system-recommended video on the first day, the data collection unit needs to be able to determine whether the user was retained on the second day. It should be able to establish a correlation between these browsing and retention behaviors.
[0075] In order for the data collected by the data acquisition unit to be used for prediction by the prediction model, a data flow unit is needed to transmit the collected data to the model prediction unit.
[0076] The model prediction unit utilizes an online real-time prediction model to extract user characteristics for each user's visit, such as gender, age, interests, and multimedia content viewed in the past 7 days, including videos and articles, and their content features. Based on these characteristics, the prediction model calculates retention rates, offering higher accuracy and efficiency than the DeepFM (Deep Factorization-Machine) model, or other deep learning model algorithms. By using the online real-time prediction model to predict user retention rates based on their current visit behavior, it's possible to obtain the retention rate after a user views a specific multimedia content, such as the estimated retention rate for the next day after watching a video.
[0077] The unified ranking unit, or model prediction unit, is just one module within the system. The obtained retention rates are sorted, and categories with higher estimated retention rates are adopted, while those with lower rates are not. For example, in a recommendation scenario, only articles and videos with high retention rates are recommended to users. Of course, this unified ranking unit can also integrate and calculate estimated retention rates with other predicted recommendation metrics from scenarios such as clicks and conversions before further ranking.
[0078] In this way, through this retention optimization process, metrics such as retention rate, click-through rate, and conversion rate can all be effectively improved, ultimately enhancing the user experience.
[0079] In some scenarios, such as e-commerce recommendation scenarios, the proposed solution can predict user retention on the second day based on user behavior within the e-commerce feed, such as the articles or videos viewed on the first day. Specifically, it first collects user browsing behavior on the first day and retention behavior on the second day. Using these browsing and retention behaviors as sample data, a DIN (Deep Interest Network) model is employed to predict retention rates. The top N articles or videos based on the predicted retention rates are then recommended to the user, thereby increasing the e-commerce platform's GMV (Gross Merchandise Volume) and click-through rate.
[0080] In some scenarios, such as news feed recommendations, the proposed solution can predict user retention on the second day based on user behavior within the news feed, such as the articles or videos viewed on the first day. Specifically, it first collects user browsing behavior on the first day and retention behavior on the second day, using these behaviors as sample data. This data is then transmitted to the model prediction unit for retention rate prediction. Based on the predicted retention rate, the top N articles or videos are ranked and recommended to the user, thereby improving the user market and user experience of the news feed.
[0081] In some scenarios, other models can also be used, such as XGBoost, or gradient boosting decision tree model, to predict retention rates for data with discrete characteristics. Specifically, the gradient boosting decision tree model can characterize the correspondence between historical behavioral data and retention rates over a predetermined future time period.
[0082] In some scenarios, the data stream of the prediction model can also be processed using streaming computation, thereby achieving online data stream output. For example, by processing online retained data through streaming computation, the processed data can be directly applied to the online real-time prediction model.
[0083] This application also provides a multimedia content recommendation device, such as... Figure 5 As shown, the device includes:
[0084] The detection unit 51 is used to detect the user's browsing behavior of the first multimedia content in the application;
[0085] The estimation unit 52 is used to obtain the estimated retention rate of the user's re-access to the application based on the historical browsing behavior of the accessed multimedia content indicated by the historical access records; wherein the estimated retention rate is associated with the content characteristics of the first multimedia content.
[0086] Recommendation unit 53 is used to determine multimedia content to be recommended to the user, based at least on the estimated retention rate.
[0087] In a specific example of this application, the estimation unit 52 is further configured to:
[0088] Based on the historical access records, the user's historical browsing behavior of accessed multimedia content at a historical time and the retention behavior at the next time after the historical time are obtained, and retention value features that characterize the contribution of the content features of the accessed multimedia content to the retention behavior are determined.
[0089] Based on the retention value characteristics and the content characteristics of the first multimedia content, the estimated retention rate of the user who visits the application again after browsing the first multimedia content is determined.
[0090] In a specific example of this application, the estimation unit 52 is further configured to:
[0091] Determine the mapping relationship between the user's historical browsing behavior and the retention behavior at the next time point in the historical time, input the mapping relationship into the prediction model, and obtain the retention value feature, which characterizes the contribution of the content features of the accessed multimedia content to the retention behavior.
[0092] In a specific example of this application, the recommendation unit 53 is further configured to:
[0093] Obtain the estimated recommendation metrics corresponding to the user, wherein the preset recommendation metrics are other recommendation metrics other than the estimated retention rate;
[0094] The estimated retention rate is weighted with the obtained estimated recommendation index, and the multimedia content to be recommended to the user is determined based on the recommendation index obtained after weighting.
[0095] In a specific example of this application, the recommendation unit 53 is further configured to:
[0096] When there are multiple first multimedia content items, the estimated retention rates corresponding to the multiple first multimedia content items are sorted, and the multimedia content to be recommended to the user is determined based on the sorting results.
[0097] In a specific example of this application, the recommendation unit 53 is further configured to:
[0098] Once the application is detected to have been accessed by the user again, multimedia content intended for that user will be recommended to that user.
[0099] It should be noted that the descriptions of the above device embodiments are similar to the descriptions of the above methods, and have the same beneficial effects as the method embodiments, therefore, they will not be repeated. For technical details not disclosed in the device embodiments of the present invention, those skilled in the art should refer to the descriptions of the method embodiments of the present invention for understanding; for the sake of brevity, they will not be repeated here.
[0100] This application also provides a multimedia content recommendation device, including: one or more processors; a memory communicatively connected to the one or more processors; and one or more applications; wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to perform the methods described above.
[0101] In a specific example, the multimedia content recommendation device described in this application embodiment may specifically be as follows: Figure 6 The structure shown indicates that the multimedia content recommendation device includes at least a processor 61, a storage medium 62, and at least one external communication interface 63; the processor 61, storage medium 62, and external communication interface 63 are all connected via a bus 64. The processor 61 can be a microprocessor, central processing unit, digital signal processor, or programmable logic array, or other electronic components with processing capabilities. The storage medium stores computer-executable code capable of executing the methods described in any of the above embodiments. In practical applications, the detection unit 51, the estimation unit 52, and the recommendation unit 53 can all be implemented using the processor 61.
[0102] It should be noted that the descriptions of the above multimedia content recommendation device embodiments are similar to the method descriptions above, and have the same beneficial effects as the method embodiments, therefore, they will not be repeated. For technical details not disclosed in the multimedia content recommendation device embodiments of the present invention, those skilled in the art should refer to the descriptions of the method embodiments of the present invention for understanding; for the sake of brevity, they will not be repeated here.
[0103] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.
[0104] Here, a computer-readable storage medium can be any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM). Furthermore, a computer-readable storage medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optical scanning of the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.
[0105] It should be understood that those skilled in the art will recognize that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0106] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk, or an optical disk, etc.
[0107] The embodiments described above are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
Claims
1. A multimedia content recommendation method, characterized in that, The method includes: The user's browsing behavior on the first multimedia content in the application was detected; Based on the historical browsing behavior of the multimedia content accessed as indicated by the historical access records, the estimated retention rate of the user's re-access to the application is obtained; the estimated retention rate is used to represent the probability that the user will access the application again after browsing the first multimedia content. Based at least on the estimated retention rate, determine the multimedia content to be recommended to the user; the multimedia content to be recommended is to be browsed by the user when they visit the application again; The method of obtaining the estimated retention rate of a user's subsequent access to the application based on historical browsing behavior of accessed multimedia content indicated by historical access records includes: Based on the historical access records, the user's historical browsing behavior of accessed multimedia content at a historical time and the retention behavior at the next time in the historical time are obtained, and retention value features that characterize the contribution of the content features of the accessed multimedia content to the retention behavior are determined; the retention behavior includes: the user accessed the multimedia content of the application at the next time, or the user did not access the multimedia content of the application at the next time. The estimated retention rate is determined based on the retention value characteristics and the content characteristics of the first multimedia content. The step of determining the multimedia content to be recommended to the user based at least on the estimated retention rate includes: Obtain the estimated recommendation metrics corresponding to the user, which are other recommendation metrics excluding the estimated retention rate; the estimated recommendation metrics are set according to the scenario. The estimated retention rate is weighted with the obtained estimated recommendation index, and the multimedia content to be recommended to the user is determined based on the recommendation index obtained after weighting.
2. The method of claim 1, wherein, The retention value features, which characterize the contribution of content features of the accessed multimedia content to retention behavior, include: Determine the mapping relationship between the user's historical browsing behavior and the retention behavior at the next time point of the historical time, input the mapping relationship into the prediction model, and obtain the retention value feature, which characterizes the contribution of the content features of the accessed multimedia content to the retention behavior.
3. The method of claim 1, wherein, The step of determining the multimedia content to be recommended to the user, based at least on the estimated retention rate, includes: When there are multiple first multimedia content items, the estimated retention rates corresponding to the multiple first multimedia content items are sorted, and the multimedia content to be recommended to the user is determined based on the sorting results.
4. The method of claim 1, wherein, The method further includes: Once the application is detected to have been accessed by the user again, multimedia content intended for that user will be recommended to that user.
5. A multimedia content recommendation apparatus, characterized by comprising: The device includes: The detection unit is used to detect the user's browsing behavior on the first multimedia content in the application; The estimation unit is used to obtain the estimated retention rate of the user's re-access to the application based on the historical browsing behavior of the accessed multimedia content indicated by the historical access records; the estimated retention rate is used to represent the probability that the user will access the application again after browsing the first multimedia content. The recommendation unit is used to determine multimedia content to be recommended to the user, based at least on the estimated retention rate; the multimedia content to be recommended is used by the user to browse when the user visits the application again; Specifically, the prediction unit is used for: Based on the historical access records, the user's historical browsing behavior of accessed multimedia content at a historical time and the retention behavior at the next time in the historical time are obtained, and retention value features that characterize the contribution of the content features of the accessed multimedia content to the retention behavior are determined; the retention behavior includes: the user accessed the multimedia content of the application at the next time, or the user did not access the multimedia content of the application at the next time. Based on the retention value characteristics and the content characteristics of the first multimedia content, the estimated retention rate is determined. Specifically, the recommendation unit is used for: Obtain the estimated recommendation metrics corresponding to the user, which are other recommendation metrics excluding the estimated retention rate; the estimated recommendation metrics are set according to the scenario. The estimated retention rate is weighted with the obtained estimated recommendation index, and the multimedia content to be recommended to the user is determined based on the recommendation index obtained after weighting.
6. A multimedia content recommendation device, characterized by, include: One or more processors; A memory that is communicatively connected to the one or more processors; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to perform the method of any one of claims 1 to 4.
7. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.