Method and apparatus for recommending a video-on-demand scheme

Abstract

The application provides a video payment scheme recommendation method and device, and relates to the technical fields of artificial intelligence, recommendation algorithm, deep learning and the like. The method comprises the following steps: obtaining a target data set associated with a video currently played by a terminal device; encoding each target data in the target data set according to the data type of the target data to obtain a feature vector set; calculating a play duration estimation model and a payment conversion rate estimation model according to the feature vector set to obtain a plurality of candidate recommended durations and a payment conversion rate corresponding to each candidate recommended duration; and generating a target video payment scheme to be recommended corresponding to the terminal device according to the plurality of candidate recommended durations and the payment conversion rate corresponding to each candidate recommended duration. In the application, the video payment scheme can be flexibly recommended, the personalized video payment mode can be dynamically recommended for the user, and the user can be provided with a high-quality viewing experience.

Description

Technical Field

[0001] This application relates to the technical fields of artificial intelligence, recommendation algorithms, and deep learning, and in particular to a recommendation method and apparatus for a video payment scheme. Background Technology

[0002] In related technologies, video payment schemes typically involve users pre-purchasing viewing time, choosing any amount of playback time they wish to buy. The price is calculated based on the actual viewing time and the total video duration. Once the user's purchased playback time is used up, they need to purchase more to continue watching. Alternatively, different trial viewing periods can be set based on user information, prompting the user to purchase the entire video after exceeding the limit. This scheme requires users to frequently trigger time purchase and payment operations during video viewing, resulting in frequent page redirects and continuous interruptions to the viewing process. This can easily cause users to miss key moments in the video, damaging their viewing experience.

[0003] Therefore, how to flexibly recommend video payment options, dynamically recommend personalized video payment methods to users, improve conversion efficiency and reduce revenue risks, while avoiding recommending extended playback time beyond the user's actual viewing intention, and providing users with a high-quality viewing experience, has become one of the important research directions. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art. The technical solution of this application is as follows:

[0005] The first aspect of this application proposes a method for recommending video payment schemes, including: Obtain the target dataset associated with the video currently playing on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video. Based on the data type of the target data, each target data item in the target dataset is encoded to obtain a set of feature vectors; Based on the feature vector set, the playback duration prediction model and the paid conversion rate prediction model are called to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration. Based on the recommended duration of multiple candidates and the corresponding paid conversion rate for each candidate recommended duration, a target video payment plan to be recommended for the terminal device is generated.

[0006] A second aspect of this application provides a device for recommending video payment schemes, comprising: The data acquisition module is used to acquire the target dataset associated with the video currently being played on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video. The feature extraction module is used to encode each target data item in the target dataset according to the data type of the target data, and obtain a set of feature vectors; The model calculation module is used to call the playback duration prediction model and the paid conversion rate prediction model based on the feature vector set to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration. The scheme generation module is used to generate a target video payment scheme to be recommended for the terminal device based on multiple candidate recommendation durations and the corresponding paid conversion rate for each candidate recommendation duration.

[0007] A third aspect of this application provides an electronic device comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the instructions to implement a recommended method for a video payment scheme as provided in the first aspect of this application.

[0008] A fourth aspect of this application provides a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a recommended method for a video payment scheme provided in the first aspect of this application.

[0009] In this embodiment, each target data item in the target dataset is encoded according to the data type of the target data to obtain a set of feature vectors. Based on the set of feature vectors, a playback duration prediction model and a payment conversion rate prediction model are called to calculate and obtain multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration. Based on the multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration, a target video payment scheme to be recommended for the terminal device is generated. This application can flexibly recommend video payment schemes, realize the dynamic recommendation of personalized video payment methods for users, improve conversion efficiency, reduce revenue risk, and avoid recommending playback durations that exceed the user's actual viewing intention, thus providing users with a high-quality viewing experience.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0011] Figure 1 This is a flowchart of a video payment scheme recommendation method according to an embodiment of this application; Figure 2 This is a flowchart of a video payment scheme recommendation method according to an embodiment of this application; Figure 3 This is a schematic diagram of the model structure of one embodiment of this application; Figure 4This is a flowchart of a video payment scheme recommendation method according to an embodiment of this application; Figure 5 This is a structural block diagram of a recommendation device for a video payment scheme according to an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0012] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0013] The acquisition, storage, and application of information and data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0014] The examples in this application may involve user data, data acquisition, and / or use. All of these aspects comply with relevant laws, regulations, and provisions. In the examples of this application, all data collection, acquisition, processing, manipulation, forwarding, and use are conducted with the user's knowledge and confirmation. Accordingly, in implementing each example, the type, scope of use, and usage scenarios of any data or information that may be involved should be communicated to the user and their authorization obtained through appropriate means in accordance with relevant laws and regulations. The specific methods of notification and / or authorization can vary depending on the actual situation and application scenario, and the scope of the solution is not limited in this regard.

[0015] In this manual and the sample solutions, any processing of personal information will be conducted only on a legal basis (such as with the consent of the data subject or as necessary for the performance of a contract) and will only be carried out within the scope stipulated or agreed upon. A user's refusal to process personal information beyond what is necessary for basic functions will not affect the user's use of basic functions.

[0016] The following description, in conjunction with the accompanying drawings, describes a recommended method and apparatus for a video payment scheme according to embodiments of this application.

[0017] Figure 1 This is a flowchart of a video payment scheme recommendation method according to one embodiment of this application, such as... Figure 1 As shown, the method includes the following steps: S101, obtain the target dataset associated with the video currently being played on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video.

[0018] The target dataset includes video content data, real-time video popularity data, real-time video competition data, and target object profiles and behavioral data; This application relates to video playback scenarios, such as user behavior analysis and personalized pricing recommendations in live streaming scenarios. Therefore, data collection, preprocessing, and feature extraction are fundamental to the system's operation. The target dataset mainly includes platform users' viewing records, payment behavior, competition content information, real-time competition status, and popularity data.

[0019] Taking the current video as an example of a live match, the target dataset can include the following data: Target audience profiles and behavioral data include target audience profile data and target audience behavioral data. Target audience profile data includes, for example, user gender, age, region, device type, viewing preferences, and willingness to pay; target audience behavioral data includes, for example, user actions such as booking, playing, sharing, liking, and paying. Video content data includes, for example, the type of match (such as football or basketball), the name of the event, the participating teams, key athletes, and the importance of the match (such as whether it is a key match in the league). In other implementations, it may also include audio information, subtitles, tags and keywords, visual elements, subject matter classification, and content theme. Real-time video match data includes information such as score difference, number of red and yellow cards, and player status. This data can also reflect the suspense level of the current match, for example, by pre-dividing it into M levels, where 0 represents virtually no suspense and M-1 represents the highest suspense. Due to the wide variety of sports, more prior knowledge needs to be incorporated. For example, in sports with only two participants (football, basketball, etc.), the score difference can be mapped to the suspense level; in sports with multiple participants, the suspense level can be calculated using the point difference between the top two teams. Real-time video popularity data, such as those calculated using metrics like real-time views, likes, comments, and shares within the platform, reflects the current popularity level of the competition.

[0020] In some implementations, the target dataset can be collected by the terminal device, and a paid recommendation request can be generated based on the target dataset and sent to the server. In some implementations, the target dataset can also be obtained directly by the server; this application does not limit this. In some implementations, the purchased duration of the video on the terminal device is obtained. In response to the difference between the purchased duration and the current playback duration being less than a preset duration threshold, the target dataset associated with the video currently being played on the terminal device is retrieved. Taking a preset duration threshold of 30 seconds as an example, during the first playback, a trial viewing time is provided. When the client's trial viewing time is about to end in 30 seconds, the client sends a paid recommendation request to the server to instruct the server to return the optimal video payment plan for the next playback.

[0021] In some implementations, the target object may be a user associated with the terminal device; in other implementations, it may be other entities or virtual entities, and this application does not limit this.

[0022] S102, based on the data type of the target data, encode each target data item in the target dataset to obtain a set of feature vectors.

[0023] Specifically, based on the data type of the target data, each target data item in the target dataset is encoded to obtain a set of feature vectors, including: performing one-hot encoding on the target data of discrete data type in the target dataset, and discretizing and one-hot encoding on the target data of continuous data type in the target dataset to obtain a set of feature vectors.

[0024] In other words, in this embodiment, discrete features are encoded using one-hot encoding. Examples include user ID, competition ID, user gender, competition type, and event name. For continuous features, they are first discretized and then one-hot encoded. Examples include user age, number of views, number of likes, and number of comments. The following is a special explanation of time-related features. All time-related features in this application need to be converted from absolute values ​​(continuous) to relative values ​​(discrete). The conversion method is as follows: Divide the estimated total match time T into N equal parts, and denote the duration of each part as a unit of time. N is an empirical value, or a hyperparameter.

[0025] any time interval The number of time units will be represented by K. , Ceil This represents the floor function.

[0026] At any time after the start of the match t (Current moment) can be represented as the time remaining until the start of the match. t A time interval of 0, that is: .

[0027] S103, based on the feature vector set, calls the playback duration prediction model and the paid conversion rate prediction model to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration.

[0028] Based on the feature vector set, the playback duration prediction model and the paid conversion rate prediction model are invoked respectively. The playback duration prediction model, based on the feature vector set, infers the duration of video content playback for the target object from the current moment; based on the duration and the configured unit time interval, multiple candidate recommendation durations are generated. The paid conversion rate prediction model, based on the feature vector set and each candidate recommendation duration, infers the paid conversion rate corresponding to each candidate recommendation duration. By aggregating and sorting different candidate recommendation durations, multiple key duration nodes and their corresponding conversion rate estimates are obtained, forming a complete candidate recommendation duration-paid conversion rate mapping set for subsequent recommendation decisions.

[0029] The unit time interval can be set as a hyperparameter or dynamically configured based on prior knowledge. The former has a fixed value for easy optimization, while the latter can adapt to data distribution and improve the rationality of time segmentation and conversion rate estimation.

[0030] In the embodiments of this application, the playback duration prediction model and the paid conversion rate prediction model can be deep neural networks (DNNs), or other networks or models that can achieve the same goal as this application, such as multilayer perceptrons.

[0031] S104. Based on the recommended duration of multiple candidates and the corresponding paid conversion rate of each candidate recommended duration, generate the target video payment plan to be recommended for the terminal device.

[0032] In this embodiment of the disclosure, based on the acquired multiple candidate recommendation durations and their corresponding paid conversion rates, a target video payment scheme to be recommended for the terminal device can be further generated.

[0033] Specifically, firstly, based on the user's current feature vector, each candidate duration-conversion rate pair is comprehensively evaluated. For example, for shorter and longer durations, the conversion rate and expected revenue are considered respectively. Next, a multi-objective optimization strategy can be adopted to maximize the expected playback duration while considering the paid conversion rate, or a conversion rate threshold can be set to filter high-potential options. Finally, considering the display constraints of the terminal device (such as interface layout and the number of recommendation slots), one or more optimal video payment options are selected and sorted in descending order of conversion rate or revenue to form a personalized recommendation list of target video payment options for real-time display and delivery on the terminal.

[0034] In this embodiment, each target data item in the target dataset is encoded according to the data type of the target data to obtain a set of feature vectors. Based on the set of feature vectors, a playback duration prediction model and a payment conversion rate prediction model are called to calculate and obtain multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration. Based on the multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration, a target video payment scheme to be recommended for the terminal device is generated. This application can flexibly recommend video payment schemes, realize the dynamic recommendation of personalized video payment methods for users, improve conversion efficiency, reduce revenue risk, and avoid recommending playback durations that exceed the user's actual viewing intention, thus providing users with a high-quality viewing experience.

[0035] Figure 2 This is a flowchart of a video payment scheme recommendation method according to one embodiment of this application, such as... Figure 2 As shown, the method includes the following steps: S201, Obtain the target dataset associated with the video currently playing on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video.

[0036] S202, based on the data type of the target data, encode each target data item in the target dataset to obtain a set of feature vectors.

[0037] For a description of steps S201 to S202, please refer to the relevant content in the above embodiments, which will not be repeated here.

[0038] S203, invoke the playback duration prediction model, and infer from the feature vector set to predict the duration of video content played by the target object from the current moment.

[0039] The playback duration prediction model and the paid conversion rate prediction model have the same model structure, which includes a control network and a main network. The main network has... h The first layer of the neural network controls the network to have h The second layer of the neural network.

[0040] Among them, the l The output of the second layer of the neural network is the first layer. l The input to the second neural network layer +1. (The first...) l The output of the first layer of the neural network and the output of the second layer l After performing bitwise vector multiplication in the second layer of the neural network, it becomes the first... l +1 layer of the second neural network.

[0041] in, h It is an integer greater than 2. l Values ​​range from 1 to h，The playback duration prediction model and the paid conversion rate prediction model were obtained by training different sample data.

[0042] For example, the target dataset includes video content data, real-time video popularity data, real-time video match data, and target object profile and behavior data. Each target data item in the target dataset is encoded according to its data type. Specifically, encoding the video content data yields a video content feature vector; encoding the real-time video popularity data yields a video popularity feature vector; encoding the real-time video match data yields a match status feature vector; and encoding the target object profile and behavior data yields a target object feature vector. Optionally, to further improve data accuracy, a current time vector can also be obtained based on the current moment, i.e., the time interval since the start of the match.

[0043] Optionally, the feature vector set includes video content feature vectors, video popularity feature vectors, competition status feature vectors, target object feature vectors, and the current time vector.

[0044] In the playback duration prediction model, the input of the first neural network in the first layer is a set of feature vectors, and the input of the second neural network in the first layer is a video popularity feature vector, a competition status feature vector, and a current playback duration vector.

[0045] Taking a live sports broadcast as an example, this demonstrates how to create a playback duration prediction model. g The model's objective is to predict the duration from the current moment when a user can continuously play live stream content. . f u Represents the feature vector of the target object. f i Represents the feature vector of video content. f h Represents the video popularity feature vector. f s Represents the feature vector of the game situation. k This represents the current time vector (i.e., the relative value with respect to the start time of the match).

[0046] DWNN() represents the model structure of this application, such as Figure 3 As shown, the DWNN model structure is mainly divided into two parts, including the main network L and the control network M.

[0047] Among them, control network M Is with L Neural networks with identical shapes, their first... l-1 Layer output As the first l The input of the layer is used to obtain the first layer.l The output of the layer is ; The first term of control network M is represented by the second term. l A multilayer neural network, with the first layer input being... ; concat ( ) indicates splicing.

[0048] Among them, the main network L The main structure is h Layered neural network, the main network L's first layer l-1 Layer output Main network L l The input of the layer is Main network L l The output of the layer is, , This represents vector bitwise multiplication. Represents the first... l Layered neural network; the first layer input of the main network L is ; DWNN is a model structure specifically designed for video playback scenarios. In traditional multi-layer neural networks, the influence of original features on the final result gradually weakens as the number of network layers increases. In live streaming scenarios, dynamic features (real-time popularity, competition status features, current time, etc.) should have a significant impact on the result. Therefore, a separate control network is built for these features. The shape of the control network is exactly the same as that of the main network, with the aim of amplifying the influence of real-time features on the final result. The output of each neuron in the main task must be adjusted by the output of the control network in the same layer before it can be input into the next layer. The effect of this approach is that even small changes in real-time features can lead to significant changes in the result.

[0049] During the training process of the playback duration prediction model, in order to construct training samples, the playback behavior of all users of the video app is collected. Each playback behavior is split into multiple samples according to time units. For example, if a user plays for K time units from the start of the game, then it will be divided into K samples. For instance, when... k=0 At that time, the video popularity characteristics were The characteristics of the competition are The actual playback duration is ;when k=1 At that time, the video popularity characteristics were The characteristics of the competition are The actual playback duration is ; and so on, when k=K-1 At that time, the video popularity characteristics were The characteristics of the competition are The actual playback duration is Playback duration estimation is a regression task, and mean squared error can be used. mse Mean absolute error mae Common loss functions are used to construct and predict playback duration. Compared to actual playback duration The model training is completed by minimizing the loss between the two gradients and performing backward gradient updates.

[0050] S204 generates multiple candidate recommended durations based on the duration and the configured unit time interval.

[0051] Based on the determined duration and configured unit time intervals, multiple candidate recommendation durations can be generated. Optionally, the unit time interval is an adjustable granularity parameter, which can be set to a fixed value (e.g., 5 minutes, 10 minutes) or dynamically adjusted based on prior knowledge (e.g., larger intervals in the early stages, smaller intervals in the later stages). Different durations correspond to different user investment costs and expected willingness to pay, providing diverse input options for subsequent calls to the paid conversion rate prediction model, facilitating the selection of the optimal recommendation scheme.

[0052] In some implementations, the time unit for the user to continue watching is obtained based on the duration. quantity This means that if the system recommends a user to purchase more than K trial viewing sessions... The probability of successful conversion will decrease significantly. Therefore, this application generates multiple candidate recommendation durations at unit time intervals based on the estimated number of user playback time units K, that is, the candidate recommendation durations d take values ​​of (1, 2, 3...K).

[0053] S205 invokes the paid conversion rate prediction model, and infers the paid conversion rate corresponding to each candidate recommendation duration based on the feature vector set and the recommendation duration of each candidate.

[0054] In the paid conversion rate prediction model, the input of the first neural network in the first layer is the set of feature vectors and the candidate recommendation duration, while the input of the second neural network in the first layer is the video popularity feature vector, the competition status feature vector, and the current playback duration vector.

[0055] The paid conversion rate prediction model in this application aims to predict the probability that a user will pay for a specified recommended playback duration; in other words, it predicts the paid conversion rate corresponding to each candidate recommendation duration.

[0056] Optionally, the paid conversion rate prediction model uses the same DWNN network structure as the playback duration prediction model, i.e.: . d The vector representing the recommended playback duration for the user is generated based on the candidate recommendation duration and will be compared with... A total of [number] inputs were made to the main network.

[0057] During the training process of the paid conversion rate prediction model, training samples are constructed using user payment prompt exposure and purchase events. When a user sees the paid continuation prompt and completes the renewal, it is considered a positive sample (label=1), and not paying is considered a negative sample (label=0). Paid conversion prediction is a classic binary classification task. By constructing the cross-entropy loss between the predicted value y and the true label, the model parameters can be updated.

[0058] S206: Based on the recommended duration of multiple candidates and the corresponding paid conversion rate for each recommended duration, generate the target video payment plan to be recommended for the terminal device.

[0059] For a description of step S206, please refer to the relevant content in the above embodiments, which will not be repeated here.

[0060] This application calls a playback duration prediction model, which infers based on a set of feature vectors to predict the duration of video content playback for the target object from the current moment. Based on the duration and the configured unit time interval, multiple candidate recommendation durations are generated. Then, a payment conversion rate prediction model is called, which infers based on the set of feature vectors and each candidate recommendation duration to predict the payment conversion rate corresponding to each candidate recommendation duration. This enables personalized segmented payment for video playback, which balances user interests and video revenue for the terminal or video playback application, providing users with a high-quality viewing experience.

[0061] Figure 4 This is a flowchart of a video payment scheme recommendation method according to one embodiment of this application, such as... Figure 4 As shown, the method includes the following steps: S401, obtain the target dataset associated with the video currently playing on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video.

[0062] S402, based on the data type of the target data, encode each target data item in the target dataset to obtain a set of feature vectors.

[0063] S403, based on the feature vector set, calls the playback duration prediction model and the paid conversion rate prediction model to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration.

[0064] For a description of steps S401 to S403, please refer to the relevant content in the above embodiments, which will not be repeated here.

[0065] S404: Calculate the expected revenue for each candidate recommendation duration based on the total video price, price fluctuation coefficient, and paid conversion rate. Then, take the candidate recommendation duration with the maximum expected revenue as the target recommendation duration. The total video price is the price corresponding to the total video duration.

[0066] Taking the current video as a live match broadcast as an example, let's assume the user needs to pay a certain price. priced With candidate recommendation duration d Proportional, candidate recommendation time is d Expected benefits of the plan 。 Indicates by d The candidate recommendation price is calculated based on the time percentage. N The number of time units into which the competition is divided. The total price is the price for the entire match, meaning the price for purchasing the entire match at once. This indicates the price fluctuation factor for pay-per-hour plans; the unit price is usually slightly higher when using this method. The common value range is 1.2 to 1.5. The price fluctuation coefficient can be set as a hyperparameter or dynamically configured based on prior knowledge. This indicates the paid conversion rate.

[0067] S405 determines the remaining playback time of the video based on the difference between the total video duration and the current playback time, and obtains the cumulative payment price for the video from the terminal device.

[0068] Based on the total video duration and the current playback time, the difference between the two is first calculated to accurately determine the remaining playback time. For example, if the total video duration is 120 minutes and 45 minutes have been played, then the remaining playback time is 75 minutes. This remaining time reflects the time cost the user still needs to invest and is an important basis for assessing renewal intentions or recommending better plans.

[0069] Simultaneously, it's necessary to obtain the cumulative paid price for the video on the terminal device, i.e., the total amount actually paid by the user from the start of viewing to the present, including single purchases, subscription sharing, or unlocking specific segments. The cumulative paid price reflects the economic cost already incurred by the user, and combined with the remaining playback time, provides data support for generating a target video payment plan. Based on these two pieces of data, recommendation schemes can be dynamically determined, thereby achieving personalized and cost-conscious video payment recommendations.

[0070] S406 generates a target video payment plan based on the target recommended duration, remaining video playback time, cumulative paid price, and total video price.

[0071] If the remaining playback time of the video is less than or equal to the target recommended duration, the target video payment plan is determined to be the total recommended video duration, and the required payment price is the difference between the total video price and the cumulative payment price.

[0072] If the remaining playback time of the video exceeds the target recommended duration, obtain the candidate recommended prices corresponding to the target recommended duration. If the sum of the candidate recommended prices and the cumulative paid price is greater than or equal to the total video price, determine the target video payment plan as recommending the purchase of the total video duration, and the required payment price is the difference between the total video price and the cumulative paid price. Alternatively, if the sum of the candidate recommended prices and the cumulative paid price is less than the total video price, determine the target video payment plan as recommending the purchase of the target recommended duration, and the required payment price is the candidate recommended prices.

[0073] Taking the current video as an example of a live match broadcast, let's assume that the recommended duration... At that time, expected return If the maximum value is selected, the target recommendation duration is determined. When determining the target recommendation duration Going forward, the target payment plan will be generated according to the following rules: If the remaining time of the match is less than or equal to the user's renewed subscription time, i.e., Kk <= d If the user chooses to purchase the total video duration, the recommended purchase price is [price missing]. ; This indicates the total cost.

[0074] If the target recommendation duration corresponds to the candidate recommendation price The sum of the cumulative paid prices is greater than or equal to the total video price, i.e. If so, the user is recommended to purchase the entire match directly, with the requested price. ; If neither of the above two conditions applies, then it is recommended to use The price is the target recommended duration, at which point the target recommended duration is... Alternatively, we recommend purchasing the entire match directly; the required price is [price missing]. .

[0075] This application supports continuous paid viewing for users. Users do not need to purchase the entire video at once; they can preview it first and then choose to extend their viewing time by paying before the current preview ends. This process can continue repeatedly until the competition concludes. This application can dynamically recommend personalized video payment methods to users, improving conversion efficiency, reducing revenue risk, and avoiding recommending extended viewing times beyond the user's actual viewing intention, thus providing a high-quality viewing experience.

[0076] For example, this application proposes a video payment scheme recommendation system for video playback scenarios. The system mainly consists of a data acquisition and processing module, a payment scheme recommendation service, and a user continuation playback interaction interface. These three components work together to form a complete closed loop for video continuation playback services. The data acquisition and processing module is responsible for collecting the target dataset and performing feature processing according to model requirements. The payment scheme generation service uses a playback duration prediction model and a payment conversion rate prediction model to calculate the optimal continuation playback duration and price in real time, generating a target video payment scheme to be recommended and sending it to the user continuation playback interaction interface. The user continuation playback interaction interface is responsible for direct interaction with the user, monitoring the user's current live playback progress in real time. If the user's purchased playback duration is less than a preset duration threshold, it requests the current target video payment scheme from the payment scheme recommendation service, prompts the user with details of the continuation playback scheme based on the returned results, and guides the user to pay.

[0077] The user resume playback interface is primarily responsible for monitoring the user's playback status, obtaining recommended resume playback options, prompting the user with resume playback information, and executing the resume playback operation. Taking a live match video as an example with a preset duration threshold of 30 seconds, the complete workflow is as follows: S1: The user watches the live stream normally. When the match ends, step S7 is executed. When there are 30 seconds left before the end of the current preview time, step S2 is executed. S2: Request the user's next renewal duration and payment amount from the paid plan recommendation service. If the user has not enabled the automatic renewal function, proceed to step S3; otherwise, proceed to step S4. S3: A text prompt will pop up below the live stream: "The current preview time is xx seconds remaining. You can pay xx yuan to continue watching [xx minutes or pay xx yuan to continue watching] all remaining matches" (the text in brackets [ ] may or may not be displayed depending on the actual situation). "xx minutes" and "all remaining matches" are clickable links. If the user clicks the link, proceed to step S4; otherwise, proceed to step S7. S4: If the user has not enabled password-free payment, redirect to the payment interface and provide the user with a QR code payment option; otherwise, display the text "Confirm payment of xx yuan [extend playback by xx minutes / purchase the entire match]?" below the video. If the user completes the QR code payment or confirms the payment on the live stream interface, proceed to step S6; otherwise, return to step S3. S5: A text prompt will pop up below the live stream: "The current preview time is xx seconds remaining. You have activated the auto-resume function and will soon pay xx yuan to continue playing for xx minutes / remaining match. Click here to cancel auto-resume." The "Cancel" link is clickable. If the user clicks it, S3 will be executed again; otherwise, step S6 will be executed after the current preview phase ends. S6: A text message will pop up below the video stating "Paid xx yuan to extend playback time by xx minutes" or "Paid xx yuan to purchase the entire match." The system will automatically extend the playback time for the user and send a payment success SMS notification. If the user has not activated the auto-resume function, they will be prompted to do so. Continue to step S1.

[0078] S7: User exits playback.

[0079] This application enables personalized recommendations for optimal continuation of viewing for users. Based on users' historical interests, payment habits, and current match preferences, combined with real-time match status and popularity, the platform dynamically recommends the next viewing duration and payment amount. The product design with automatic continuation eliminates the need for manual renewal when users select automatic continuation and enable password-free payment, preventing interruptions during crucial moments of the match and ensuring a smooth viewing experience.

[0080] Figure 5 This is a structural block diagram of a recommendation device for a video payment scheme according to an embodiment of this application, as shown below. Figure 5 As shown, the recommendation device 500 for the video payment scheme includes: Data acquisition module 510 is used to acquire the target dataset associated with the video currently being played on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video. The feature extraction module 520 is used to encode each target data in the target dataset according to the data type of the target data to obtain a set of feature vectors; The model calculation module 530 is used to call the playback duration prediction model and the paid conversion rate prediction model based on the feature vector set to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration. The scheme generation module 540 is used to generate a target video payment scheme to be recommended for the terminal device based on multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration.

[0081] In some implementations, the model calculation module 530 is also used for: The playback duration prediction model is invoked, and inference is performed based on the feature vector set to predict the duration of video content played by the target object from the current moment. Based on the duration and the configured unit time interval, generate multiple candidate recommendation durations; The paid conversion rate prediction model is invoked, and inferences are performed based on the feature vector set and the duration of each candidate recommendation to predict the paid conversion rate corresponding to each candidate recommendation duration.

[0082] In some implementations, the playback duration prediction model and the paid conversion rate prediction model have the same model structure, which includes a control network and a main network. The main network has an h-layer first neural network, and the control network has an h-layer second neural network. The output of the second neural network at layer l is the input of the second neural network at layer l+1. The output of the first neural network at layer l and the second neural network at layer l are multiplied by vector by bit, and then used as the input of the second neural network at layer l+1. This process continues until the output of the first neural network at layer h and the second neural network at layer h are multiplied by vector by bit to obtain the output of the model structure. Where h is an integer greater than 2, l takes values ​​from 1 to h, and the playback duration prediction model and the paid conversion rate prediction model are obtained by training different sample data.

[0083] In some implementations, the feature vector set includes video content feature vectors, video popularity feature vectors, match status feature vectors, target object feature vectors, and the current time vector; In the playback duration prediction model, the input of the first neural network in the first layer is a set of feature vectors, and the input of the second neural network in the first layer is a video popularity feature vector, a competition status feature vector, and a current playback duration vector. In the paid conversion rate prediction model, the input of the first neural network in the first layer is the set of feature vectors and the candidate recommendation duration, while the input of the second neural network in the first layer is the video popularity feature vector, the competition status feature vector, and the current playback duration vector.

[0084] In some implementations, the target dataset includes video content data, real-time video popularity data, real-time video match data, and target object profiles and behavioral data; The feature extraction module 520 is also used for: One-hot encoding is performed on the discrete data in the target dataset, and discretization and one-hot encoding are performed on the continuous data in the target dataset to obtain a set of feature vectors.

[0085] In some implementations, the scheme generation module 540 is further configured to: Based on the total video price, price fluctuation coefficient, and paid conversion rate, calculate the expected revenue corresponding to each candidate recommendation duration, and take the candidate recommendation duration with the maximum expected revenue as the target recommendation duration. The total video price is the price corresponding to the total video duration. The remaining playback time of the video is determined based on the difference between the total video duration and the current playback time, and the cumulative payment price for the video by the terminal device is obtained. A target video payment plan is generated based on the target recommended duration, remaining video playback time, cumulative payment price, and total video price.

[0086] In some implementations, the scheme generation module 540 is further configured to: If the remaining playback time of the video is less than or equal to the target recommended duration, the target video payment plan is determined to be the total recommended video duration, and the required payment price is the difference between the total video price and the cumulative payment price.

[0087] In some implementations, the scheme generation module 540 is further configured to: If the remaining playback time of the video is greater than the target recommendation duration, obtain the candidate recommendation price corresponding to the target recommendation duration; If the sum of the candidate recommended price and the cumulative paid price is greater than or equal to the total video price, the target video payment plan is determined to be the recommended purchase of the total video duration, and the required payment price is the difference between the total video price and the cumulative paid price; or, If the sum of the candidate recommended price and the cumulative paid price is less than the total price of the video, the target video payment plan is determined to be the recommended purchase duration, and the required payment price is the candidate recommended price.

[0088] In some implementations, the data acquisition module 510 is further configured to: Obtain the purchased duration of the video from the terminal device; If the difference between the purchased duration and the current playback duration is less than a preset duration threshold, the system will trigger the acquisition of the target dataset associated with the video currently playing on the terminal device.

[0089] In this embodiment, each target data item in the target dataset is encoded according to the data type of the target data to obtain a set of feature vectors. Based on the set of feature vectors, a playback duration prediction model and a payment conversion rate prediction model are called to calculate and obtain multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration. Based on the multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration, a target video payment scheme to be recommended for the terminal device is generated. This application can flexibly recommend video payment schemes, realize the dynamic recommendation of personalized video payment methods for users, improve conversion efficiency, reduce revenue risk, and avoid recommending playback durations that exceed the user's actual viewing intention, thus providing users with a high-quality viewing experience.

[0090] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0091] like Figure 6 As shown, the electronic device 600 includes: The memory 601 and the processor 602 are connected by a bus 603. The memory 601 stores a computer program, which, when executed by the processor 602, implements the recommended method of the video payment scheme of the present application embodiment.

[0092] Bus 603 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0093] Electronic device 600 typically includes a variety of electronic device readable media. These media can be any available media that can be accessed by electronic device 600, including volatile and non-volatile media, removable and non-removable media.

[0094] Memory 601 may also include computer system readable media in the form of volatile memory, such as random access memory (RAM) 604 and / or cache memory 605. Electronic device 600 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 606 can be used to read and write non-removable, non-volatile magnetic media (… Figure 6 Not shown; usually referred to as a "hard drive"). Although Figure 6 As not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 603 via one or more data media interfaces. Memory 601 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0095] A program / utility 608 having a set (at least one) of program modules 607 may be stored, for example, in memory 601. Such program modules 607 include, but are not limited to, an 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. Program modules 607 typically perform the functions and / or methods described in the embodiments of this application.

[0096] Electronic device 600 can also communicate with one or more external devices 609 (e.g., keyboard, pointing device, display 611, etc.), and with one or more devices that enable a user to interact with the electronic device 600, and / or with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 612. Furthermore, electronic device 600 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) through network adapter 613. Figure 6 As shown, network adapter 613 communicates with other modules of electronic device 600 via bus 603. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0097] The processor 602 executes various functional applications and data processing by running programs stored in the memory 601.

[0098] It should be noted that the implementation process and technical principles of the electronic device in this embodiment are explained in the foregoing description of the recommended method for video payment schemes in this application embodiment, and will not be repeated here.

[0099] To implement the above embodiments, this application also proposes a computer-readable storage medium.

[0100] When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the recommended method of the aforementioned video payment scheme. Optionally, the computer-readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0101] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0102] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for recommending paid video subscription plans, characterized in that, include: Obtain the target dataset associated with the video currently playing on the terminal device, wherein the target dataset includes at least the multi-dimensional target data corresponding to the target object and the video; Based on the data type of the target data, each target data item in the target dataset is encoded to obtain a set of feature vectors; Based on the set of feature vectors, the playback duration prediction model and the paid conversion rate prediction model are called to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration. Based on the multiple candidate recommendation durations and the corresponding paid conversion rate for each candidate recommendation duration, a target video payment plan to be recommended for the terminal device is generated.

2. The method according to claim 1, characterized in that, The step involves using the feature vector set to call the playback duration prediction model and the paid conversion rate prediction model to calculate and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration, including: The playback duration prediction model is invoked, and reasoning is performed based on the feature vector set to predict the duration of video content played by the target object from the current moment. Based on the duration and the configured unit time interval, multiple candidate recommendation durations are generated; The paid conversion rate prediction model is invoked, and inference is performed based on the feature vector set and the duration of each candidate recommendation to predict the paid conversion rate corresponding to each candidate recommendation duration.

3. The method according to claim 2, characterized in that, The playback duration prediction model and the paid conversion rate prediction model have the same model structure, which includes a control network and a main network. The main network has... h The first layer of the neural network, the control network having h Second layer of neural network; Among them, the l The output of the second layer of the neural network is the first layer. l +1 layer of the second neural network input; will the first l The output of the first layer of the neural network and the output of the second layer l After performing bitwise vector multiplication in the second layer of the neural network, it becomes the first... l +1 layer of the second neural network input, until the first h The output of the first layer of the neural network and the output of the second layer h The second layer of the neural network performs bitwise vector multiplication to obtain the output of the model structure. Among them, the h It is an integer greater than 2. l Values ​​range from 1 to h， The playback duration prediction model and the paid conversion rate prediction model are obtained by training different sample data.

4. The method according to claim 3, characterized in that, The feature vector set includes video content feature vectors, video popularity feature vectors, competition status feature vectors, target object feature vectors, and the current time vector; Wherein, the input of the first neural network in the first layer of the playback duration prediction model is the set of feature vectors, and the input of the second neural network in the first layer of the playback duration prediction model is the video popularity feature vector, the competition status feature vector, and the current playback duration vector; In the paid conversion rate prediction model, the input of the first neural network of the first layer is the feature vector set and the candidate recommendation duration, and the input of the second neural network of the first layer is the video popularity feature vector, the competition status feature vector and the current playback duration vector.

5. The method according to any one of claims 1-4, characterized in that, The target dataset includes video content data, real-time video popularity data, real-time video competition data, and target object profiles and behavioral data. The step of encoding each target data item in the target dataset according to the data type of the target data to obtain a set of feature vectors includes: One-hot encoding is performed on the discrete data in the target dataset, and discretization and one-hot encoding are performed on the continuous data in the target dataset to obtain the feature vector set.

6. The method according to any one of claims 1-4, characterized in that, The step of generating a target video payment plan for the terminal device based on multiple candidate recommendation durations and the corresponding payment conversion rate for each candidate recommendation duration includes: Based on the total video price, the price fluctuation coefficient, and the paid conversion rate, the expected revenue corresponding to each candidate recommendation duration is calculated, and the candidate recommendation duration with the maximum expected revenue is taken as the target recommendation duration. The total video price is the price corresponding to the total video duration. The remaining playback time of the video is determined based on the difference between the total duration of the video and the current playback time, and the cumulative payment price paid by the terminal device for the video is obtained. The target video payment plan is generated based on the target recommended duration, the remaining playback time of the video, the cumulative paid price, and the total price of the video.

7. The method according to claim 6, characterized in that, The step of generating the target video payment plan based on the target recommended duration, the remaining playback time of the video, the cumulative paid price, and the total price of the video includes: In response to the remaining playback time of the video being less than or equal to the target recommended duration, the target video payment plan is determined to be the recommended purchase of the total video duration, and the required payment price is the difference between the total video price and the cumulative payment price.

8. The method according to claim 6, characterized in that, The step of generating the target video payment plan based on the target recommended duration, the remaining playback time of the video, the cumulative paid price, and the total price of the video includes: In response to the remaining playback time of the video being greater than the target recommendation duration, the candidate recommendation price corresponding to the target recommendation duration is obtained; In response to the fact that the sum of the candidate recommended price and the cumulative paid price is greater than or equal to the total video price, the target video payment plan is determined to be the recommended purchase of the total video duration, and the required payment price is the difference between the total video price and the cumulative paid price; or, In response to the fact that the sum of the candidate recommended price and the cumulative paid price is less than the total price of the video, the target video payment plan is determined to be the recommended purchase duration, and the demand payment price is the candidate recommended price.

9. The method according to any one of claims 1-4, characterized in that, The acquisition of the target dataset associated with the video currently playing on the terminal device also includes: Obtain the purchased duration of the video on the terminal device; In response to the difference between the purchased duration and the current playback duration being less than a preset duration threshold, the system triggers the acquisition of the target dataset associated with the video currently being played on the terminal device.

10. A device for recommending video payment schemes, characterized in that, include: The data acquisition module is used to acquire the target dataset associated with the video currently being played on the terminal device. The target dataset includes at least the multi-dimensional target data corresponding to the target object and the video. The feature extraction module is used to encode each target data in the target dataset according to the data type of the target data to obtain a set of feature vectors; The model calculation module is used to call the playback duration prediction model and the paid conversion rate prediction model to calculate based on the feature vector set, and obtain multiple candidate recommendation durations and the paid conversion rate corresponding to each candidate recommendation duration; The scheme generation module is used to generate a target video payment scheme to be recommended for the terminal device based on the multiple candidate recommendation durations and the corresponding payment conversion rate of each candidate recommendation duration.

11. An electronic device, characterized in that, include: processor; Memory for storing the executable instructions of the processor; The processor is configured to execute the instructions to implement the method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method as described in any one of claims 1-9.

Images