Recommended content processing method and device, computer device, and storage medium

Abstract

The application relates to a processing method and device for recommending content, computer equipment and a storage medium. The method comprises the following steps: performing conversion estimation on first recommended content to obtain initial estimation conversion data of the first recommended content; obtaining historical conversion data and estimation conversion data corresponding to second recommended content; the second recommended content matches an attribute value of the first recommended content in a target attribute category; determining calibration information corresponding to the initial estimation conversion data according to the difference between the historical conversion data and the estimation conversion data; and adjusting the initial estimation conversion data based on the calibration information to obtain a conversion estimation result corresponding to the first recommended content. The method can effectively improve the conversion estimation accuracy of recommended content.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, computer device, and storage medium for processing recommended content. Background Technology

[0002] With the rapid development of computer technology, various intelligent recommendation technologies are becoming increasingly mature. Content recommendation is one such scenario. In content recommendation, it is usually necessary to determine the conversion data generated by the recommended content. This conversion data is typically used to measure the effectiveness of the recommended content; generally, the more successful the conversion, the better the recommendation effect.

[0003] Traditionally, conversion predictions are made based on the data of the recommended content itself. However, even highly exposed recommended content has a limited amount of data, leading to low accuracy in conversion forecasting when relying solely on the content's own data. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for processing recommended content that can effectively improve the accuracy of conversion prediction of recommended content, in order to address the above-mentioned technical problems.

[0005] A method for processing recommended content, the method comprising:

[0006] The conversion rate of the first recommended content is estimated to obtain the initial estimated conversion data of the first recommended content;

[0007] Obtain historical conversion data and estimated conversion data corresponding to the second recommended content; the second recommended content matches the attribute value of the first recommended content under the target attribute category.

[0008] Based on the difference between the historical conversion data and the estimated conversion data, calibration information corresponding to the initial estimated conversion data is determined;

[0009] The initial estimated conversion data is adjusted based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0010] A recommended content processing apparatus, the apparatus comprising:

[0011] The conversion prediction module is used to predict the conversion of the first recommended content and obtain the initial predicted conversion data of the first recommended content.

[0012] The data acquisition module is used to acquire historical conversion data and estimated conversion data corresponding to the second recommended content; the second recommended content is recommended content with the same content attributes as the first recommended content;

[0013] The calibration information determination module is used to determine the calibration information corresponding to the initial estimated conversion data based on the difference between the historical conversion data and the estimated conversion data.

[0014] The data calibration module is used to adjust the initial estimated conversion data based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0015] In one embodiment, the data acquisition module is further configured to acquire conversion data corresponding to candidate recommended content; the candidate recommended content is recommended content that matches the attribute value of the first recommended content under the target attribute category; the conversion data includes historical conversion data and estimated conversion data; from the conversion data, historical conversion data and estimated conversion data corresponding to the second recommended content are extracted; the second recommended content is recommended content among the candidate recommended content that meets the preset filtering conditions.

[0016] In one embodiment, the data acquisition module is further configured to determine candidate recommended content that matches the attribute value of the first recommended content under the target attribute category; acquire conversion data of the candidate recommended content in multiple time periods; the amount of conversion data in each time period is negatively correlated with the distance of the time period from the current time.

[0017] In one embodiment, the data acquisition module is further configured to determine the current target attribute category according to the priority order of each target attribute category when there are multiple target attribute categories; if the candidate recommended content under the current target attribute category does not meet the content extraction conditions, then select the next target attribute category as the current target attribute category; until the candidate recommended content under the current target attribute category meets the content extraction conditions, then select the second recommended content from the candidate recommended content under the current target attribute category.

[0018] In one embodiment, the calibration information determination module is configured to, when the first recommended content is in the initial exposure stage, acquire an amplification factor corresponding to the initial estimated conversion data; and determine calibration information corresponding to the initial estimated conversion data based on the difference between the historical conversion data and the estimated conversion data and the amplification factor.

[0019] In one embodiment, the calibration information determination module is used to determine the target resource consumption range corresponding to the first recommended content from a preset resource consumption range; the preset resource consumption range is pre-divided according to the difference between the actual resource consumption value and the target resource consumption value of the first recommended content; and to determine the expansion coefficient corresponding to the target resource consumption range.

[0020] In one embodiment, the calibration information determination module is used to obtain the deviation between the current actual total resource consumption and the current target total resource consumption; the current actual total resource consumption is the total amount of resources currently actually consumed by the first recommended content; the current target total resource consumption is the target total resource consumption corresponding to the first recommended content; and based on the deviation, an expansion coefficient corresponding to the target resource consumption range is determined.

[0021] In one embodiment, the calibration information determination module is used to obtain a smoothing coefficient when the first recommended content is in a fully exposed stage; the smoothing coefficient is determined based on the historical conversion data corresponding to the first recommended content; and calibration information corresponding to the initial estimated conversion data is determined based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data; wherein the smoothed historical conversion data is obtained by smoothing the historical conversion data corresponding to the second recommended content according to the smoothing coefficient; and the smoothed estimated conversion data is obtained by smoothing the estimated conversion data corresponding to the second recommended content according to the smoothing coefficient.

[0022] In one embodiment, the calibration information determination module is configured to determine resource consumption deviation information based on the target resource consumption level corresponding to the first recommended content, the historical conversion data of the second recommended content, and the estimated conversion data; adjust the historical conversion data of the second recommended content based on the resource consumption deviation information to obtain adjusted historical conversion data; and smooth the adjusted historical conversion data corresponding to the second recommended content according to the smoothing coefficient to obtain smoothed historical conversion data.

[0023] In one embodiment, the calibration information determination module is configured to extract historical conversion data and estimated conversion data under the target resource consumption level from the historical conversion data and estimated conversion data of the second recommended content according to the target resource consumption level corresponding to the first recommended content; and determine resource consumption deviation information based on the difference between the historical conversion data and estimated conversion data under the target resource consumption level and the historical conversion data and estimated conversion data of the second recommended content.

[0024] In one embodiment, the conversion prediction module is further configured to obtain content attribute information and recommendation object information corresponding to the first recommended content; extract content features from the content attribute information and object features from the recommendation object information using a trained conversion prediction model; and perform conversion prediction on the first recommended content based on the content features and the object features to obtain initial predicted conversion data for the first recommended content.

[0025] In one embodiment, the above-mentioned recommended content processing device further includes a model training module for acquiring sample content and training labels; the sample content includes historical recommended content, and the training labels are determined based on historical conversion data and historical access data corresponding to the historical recommended content; through the conversion prediction model to be trained in the current round, sample content features and sample object features corresponding to the sample content are extracted; based on the sample content features and the sample object features, the conversion of the sample content is predicted to obtain sample predicted conversion data; based on the difference between the sample predicted conversion data and the training labels, the parameters of the conversion prediction model are adjusted, the next round is taken as the current round and training continues until the training conditions are met and training stops.

[0026] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the processing method of the recommended content of the various embodiments of this application.

[0027] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the processing method of the recommended content of the various embodiments of this application.

[0028] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, it implements the steps in the processing method of the recommended content of the various embodiments of this application.

[0029] The aforementioned method, apparatus, computer equipment, and storage medium for processing recommended content involve the computer equipment performing conversion prediction on the first recommended content to obtain initial predicted conversion data. Then, it acquires historical conversion data and predicted conversion data corresponding to the second recommended content. The attribute values ​​of the second recommended content and the first recommended content are matched under the target attribute category. Based on the difference between the historical conversion data and the predicted conversion data, calibration information corresponding to the initial predicted conversion data is determined. Then, based on this calibration information, the initial predicted conversion data of the first recommended content is adjusted. This effectively utilizes historical data of the second recommended content under the same target attribute category to calibrate and correct the initial predicted conversion data of the first recommended content, reducing errors caused by directly predicting the conversion of recommended content based on its own data. This results in more accurate conversion prediction results and effectively improves the accuracy of conversion prediction for recommended content. Attached Figure Description

[0030] Figure 1This is an application environment diagram of the method for processing recommended content in one embodiment;

[0031] Figure 2 This is a schematic diagram of a content recommendation system in one embodiment;

[0032] Figure 3 This is a flowchart illustrating a method for processing recommended content in one embodiment;

[0033] Figure 4 This is a schematic diagram illustrating the process of promoting advertising information in one embodiment;

[0034] Figure 5 This is a flowchart illustrating the method for processing recommended content in another embodiment;

[0035] Figure 6 This is a flowchart illustrating the steps of model training in one embodiment;

[0036] Figure 7 This is a flowchart illustrating the method for processing recommended content in yet another embodiment;

[0037] Figure 8 This is a structural block diagram of a processing device for recommended content in one embodiment;

[0038] Figure 9 A structural block diagram of a processing apparatus for recommended content in another embodiment;

[0039] Figure 10 This is an internal structural diagram of a computer device in one embodiment;

[0040] Figure 11 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0041] 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.

[0042] The address data processing method provided in this application can be applied to computer devices. The computer device can be a terminal or a server. It is understood that the address data processing method provided in this application can be applied to a terminal, a server, or a system including both a terminal and a server, and is implemented through interaction between the terminal and the server.

[0043] The method for processing the recommended content provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. Specifically, terminal 102 can upload the acquired first recommended content to server 104, and server 104 performs conversion prediction on the acquired first recommended content to obtain initial predicted conversion data for the first recommended content. Then, server 104 obtains historical conversion data and predicted conversion data corresponding to second recommended content; the attribute values ​​of the second recommended content and the first recommended content are matched under the target attribute category. Based on the difference between the historical conversion data and the predicted conversion data, server 104 determines calibration information corresponding to the initial predicted conversion data, and then adjusts the initial predicted conversion data based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0044] The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any restrictions.

[0045] It is understood that the methods for processing recommended content in the various embodiments of this application can employ artificial intelligence, machine learning, and cloud technologies, which can effectively perform conversion prediction processing on recommended content and calibrate the initial conversion prediction data of recommended content. This results in more accurate conversion prediction results and improves the accuracy of conversion prediction for recommended content. Artificial intelligence (AI) is the theory, methods, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. Artificial intelligence studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making functions.

[0046] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning. It can be understood that the conversion prediction model used in some embodiments of this application is trained using machine learning techniques based on artificial intelligence. The conversion prediction model trained based on these machine learning techniques can more accurately predict the conversion data of recommended content.

[0047] Cloud computing refers to the delivery and usage model of IT (Internet Technology) infrastructure, meaning obtaining necessary resources in an on-demand and easily scalable manner through a network. In a broader sense, cloud computing also refers to the delivery and usage model of services, meaning obtaining necessary services in an on-demand and easily scalable manner through a network. These services can be IT and software related, internet-related, or other services. Cloud computing is a product of the convergence and development of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing. Driven by the development of the internet, real-time data streams, the diversification of connected devices, and the demands of search services, social networks, mobile commerce, and open collaboration, cloud computing has developed rapidly. Unlike previous parallel and distributed computing, the emergence of cloud computing will fundamentally revolutionize the entire internet model and enterprise management model. In some embodiments of this application, when there are a large number of business requests in the business system corresponding to the recommended content, cloud computing technology can be used to process the large number of business requests concurrently, thereby effectively improving the processing efficiency of content recommendation.

[0048] In another embodiment, the method for processing recommended content provided in this application can be applied to, for example, Figure 2 The application environment shown includes a terminal 202 and a content recommendation system 204, which can specifically be a blockchain-based business system. (Reference) Figure 2 A blockchain-based business system comprises multiple layers of nodes. Every node in the entire blockchain business system can be represented as a blockchain node, and each node can be provided as a server. The blockchain-based business system can receive content recommendation requests initiated by terminal 110 and, based on the attribute information of the recommended content in the request, distribute the content recommendation request to the corresponding blockchain nodes in the blockchain-based business system for processing.

[0049] Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying platform, a platform product service layer, and an application service layer. In the embodiments of this application, blockchain can be used to process recommended content to encrypt and verify its validity.

[0050] The platform's product service layer provides basic capabilities and implementation frameworks for typical applications. Developers can leverage these capabilities, along with the specific characteristics of their business needs, to implement blockchain-based business logic. The application service layer provides blockchain-based application services to business stakeholders. For example, it can use blockchain technology to encrypt and process recommended content provided by stakeholders.

[0051] In one embodiment, such as Figure 3 As shown, a method for processing recommended content is provided. Taking the application of this method to a computer device as an example, the computer device can specifically be a terminal or a server.

[0052] In this embodiment, the method includes the following steps:

[0053] Step S302: Perform conversion prediction on the first recommended content to obtain the initial predicted conversion data of the first recommended content.

[0054] Recommended content refers to content that needs to be recommended to the target audience. For example, recommended content can include push notifications and resource promotion information. Resource promotion information includes product information and advertising information. Push notifications refer to information that can be automatically pushed to target audiences, such as those interested in the information. Target audiences can be those interested in the information. Push notifications can include community information, video information, article information, application software information, and user information. Resource promotion information refers to information that pushes specified content to a target user audience and calculates resource consumption based on a specific resource calculation model. Resource promotion information can include product information and advertising information.

[0055] The "recommendation target" refers to the intended audience for recommending the content, such as a specific group of people to whom the content is targeted. The format of the recommended content can include at least one of the following: plain text, plain images, icons, or a combination of text and images.

[0056] Conversion can be understood as the action of a visitor completing a target conversion behavior. Conversion prediction refers to the estimation of conversion rates for recommended content, specifically conversion rate prediction. The conversion rate is the ratio of the number of times a visitor performs a corresponding target conversion behavior to the total number of visits. Target conversion behavior refers to the expected behavior towards the recommended content, such as browsing time or triggering a target action based on the content. Target behaviors include registration, forwarding, communication, and purchase completion.

[0057] The initial estimated conversion data for the first recommended content refers to the initial estimated conversion data obtained after performing conversion estimation processing on the first recommended content. Specifically, it can be the initial estimated conversion rate, where the conversion rate refers to the probability that the recommended content will be converted after being clicked.

[0058] The first recommended content includes content attribute information and recommendation target information. Content attribute information refers to the inherent attributes of the first recommended content itself, such as category attributes, affiliation attributes, and cluster attributes. Recommendation target information refers to the information of the intended audience for the first recommended content. This information can be used to indicate the targeting attributes of the recommended content, such as the target audience's gender, age, region, occupation, education level, and place of residence.

[0059] The computer device can first perform conversion prediction processing on the first recommended content. Specifically, the computer device extracts features from the first recommended content, extracting content features from content attribute information and object features from the recommended object information. Then, based on the content features and object features, the computer device performs conversion prediction on the first recommended content, thereby obtaining the initial predicted conversion data for the first recommended content.

[0060] Step S304: Obtain the historical conversion data and estimated conversion data corresponding to the second recommended content.

[0061] It can be understood that the second recommended content is different from the first recommended content. The second recommended content is content whose attribute value matches that of the first recommended content under the target attribute category. For example, the attribute category of the first recommended content matches the attribute value of the attribute category of the second recommended content under one or more target attribute categories. Here, the attribute category refers to the category of attribute information of the recommended content, and the target attribute category can refer to the category of attribute information of one or more pre-defined recommended content items. The attribute information of the recommended content may include at least one of the following: attribution attribute, cluster attribute, content category attribute, promotion site attribute, etc.

[0062] Historical conversion data refers to the conversion data generated by the historical exposure of the recommended content; that is, the conversion data of the recommended content within a certain historical period. Specifically, conversion tracking tools can be used to track the conversion of recommended content and obtain the corresponding conversion data. Historical conversion data can include the number of conversions, conversion record information, etc. Conversion record information includes conversion time, conversion site platform, and other conversion event information. Estimated conversion data refers to the estimated conversion data for the recommended content within a certain historical period.

[0063] Specifically, when the recommended content is promotional information, such as advertising information, the attribute information of the advertising information may include advertising identifiers, product identifiers, brand identifiers, advertiser identifiers, advertising categories, advertising placement time, advertising placement sites, and other information that has the content characteristics of the advertisement itself.

[0064] The computer device performs conversion estimation on the first recommended content and obtains the corresponding initial estimated conversion data. Then, it acquires the historical conversion data and estimated conversion data corresponding to the second recommended content. Specifically, the computer device can first acquire conversion data corresponding to multiple recommended content items that match the attribute values ​​of the first recommended content under the target attribute category. Then, from the conversion data corresponding to multiple recommended content items, it extracts the historical conversion data and estimated conversion data corresponding to a portion of the recommended content that meets preset filtering conditions. The recommended content that meets the preset filtering conditions is then designated as the second recommended content.

[0065] In another embodiment, the computer device may also determine a second recommended content that matches the attribute value of the first recommended content under the target attribute category based on the target attribute category, and then obtain the historical conversion data and estimated conversion data corresponding to the second recommended content.

[0066] In another embodiment, the computer device can also determine multiple candidate recommended contents that match the attribute values ​​of the first recommended content under the target attribute category according to the target attribute category, and then extract a portion of the recommended contents that meet the preset filtering conditions from the multiple candidate recommended contents as the second recommended contents, and then obtain the historical conversion data and estimated conversion data corresponding to the second recommended contents.

[0067] Step S306: Based on the difference between historical conversion data and estimated conversion data, determine the calibration information corresponding to the initial estimated conversion data.

[0068] Understandably, the difference between historical conversion data and projected conversion data refers to the deviation between the projected conversion and the actual conversion for the recommended content. Calibration information refers to corrective information used to reduce the error in the conversion prediction of recommended content; specifically, it can be calibration coefficients or correction coefficients.

[0069] After the computer device obtains the historical conversion data and estimated conversion data corresponding to the second recommended content, it then determines the difference between the historical conversion data and the estimated conversion data based on the historical conversion data and the estimated conversion data corresponding to the second recommended content, so as to determine the calibration information corresponding to the initial estimated conversion data of the first recommended content based on the difference between the historical conversion data and the estimated conversion data.

[0070] Step S308: Adjust the initial estimated conversion data based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0071] It is understandable that the conversion prediction result corresponding to the first recommended content is the target conversion prediction result after calibrating the initial conversion prediction data.

[0072] The computer equipment determines calibration information corresponding to the initial estimated conversion data based on the difference between historical conversion data and estimated conversion data corresponding to the second recommended content. This calibration information is then used to adjust and correct the initial estimated conversion data for the first recommended content, making the conversion prediction for the first recommended content more accurate, thereby obtaining a more precise conversion prediction result.

[0073] The difference between historical conversion data and estimated conversion data can reflect the conversion deviation between the actual conversion data and the estimated conversion data of the second recommended content. By using the conversion deviation of the second recommended content, which matches the attribute value of the first recommended content under the target attribute category, as the calibration information corresponding to the initial estimated conversion data of the first recommended content, the initial estimated conversion data of the first recommended content can be calibrated by referring to the conversion deviation of the second recommended content.

[0074] In one embodiment, after obtaining the conversion prediction result corresponding to the first recommended content, the computer device can further configure the recommendation attribute information of the first recommended content according to the conversion prediction result, or adjust the recommendation attribute information of the first recommended content according to the conversion prediction result, so as to recommend the first recommended content according to the recommendation attribute information of the first recommended content.

[0075] For example, based on conversion forecasts and the estimated resource consumption of the top recommended content, the promotional resource consumption of the top recommended content can be determined. Then, the top recommended content can be promoted based on this resource consumption. Specifically, bidding can be used to promote the top recommended content based on its resource consumption to determine the target promotion sites, thereby recommending the top recommended content on those target sites.

[0076] Taking recommended content as advertising information as an example, the promotion process of advertising information can be as follows: Figure 4 As shown, refer to Figure 4The advertising process involves four stages: exposure, click, conversion, and payment. The billing point (BNP) is how the advertising platform charges for ads, while the bid point is how advertisers bid. If the BNP is in the exposure stage, the platform charges based on the number of ad impressions; if it's in the click stage, it charges based on the number of clicks, and so on. If the bid point is in the exposure stage, the advertiser bids based on the number of ad impressions; if it's in the click stage, it bids based on the number of clicks. If both the BNP and bid points are concentrated in the exposure stage, the ad is a CPM (Cost Per Mille) ad; if they are concentrated in the click stage, it's a CPC (Cost Per Click) ad; and if they are concentrated in the conversion stage, it's a CPA (Cost Per Action) ad. If the billing point is at the click stage and the bidding point is at the conversion stage, it is an OCPA (Optimized Cost Per Action) ad, which means that the price is based on the actual conversion effect of the ad, and can be determined by estimating PCVR (Predicted Conversion Rate).

[0077] In one embodiment, the conversion prediction result can also be used to rank and recommend the first recommended content when further recommending it. It can also be used to determine whether to display the content to the user based on the conversion prediction result.

[0078] In the above-mentioned method for processing recommended content, the computer device performs conversion prediction on the first recommended content to obtain initial predicted conversion data. Then, it acquires historical conversion data and predicted conversion data corresponding to the second recommended content. The attribute values ​​of the second recommended content and the first recommended content are matched under the target attribute category. Based on the difference between the historical conversion data and the predicted conversion data, calibration information corresponding to the initial predicted conversion data is determined. Then, based on this calibration information, the initial predicted conversion data of the first recommended content is adjusted. This effectively utilizes historical data of the second recommended content under the same target attribute category to calibrate and correct the initial predicted conversion data of the first recommended content, reducing errors caused by directly predicting the conversion of recommended content based on its own data. This results in a more accurate conversion prediction result, thereby effectively improving the accuracy of conversion prediction for recommended content.

[0079] In one embodiment, obtaining historical conversion data and estimated conversion data corresponding to the second recommended content includes: obtaining conversion data corresponding to candidate recommended content; candidate recommended content is recommended content that matches the attribute value of the first recommended content under the target attribute category; the conversion data includes historical conversion data and estimated conversion data; extracting historical conversion data and estimated conversion data corresponding to the second recommended content from the conversion data; the second recommended content is recommended content among the candidate recommended content that meets preset filtering conditions.

[0080] Among them, candidate recommended content refers to multiple recommended contents that match the attribute values ​​of the first recommended content under multiple target attribute categories, and is used to extract the second recommended content from the multiple recommended contents.

[0081] Specifically, the computer device can obtain candidate recommended content that matches the attribute values ​​of multiple target attribute categories based on the attribute values ​​of the first recommended content. Then, the computer device obtains the corresponding conversion data for the candidate recommended content, wherein the conversion data includes historical conversion data and estimated conversion data for each candidate recommended content.

[0082] Specifically, the computer device can obtain click event records and conversion event records of the candidate recommended content over a historical period. Click event records include information such as click time, number of clicks, and clicked site. Conversion event records include information such as conversion time, number of conversions, and conversion site.

[0083] Then, the computer equipment filters the historical conversion data and estimated conversion data of the second recommended content that meet the preset filtering conditions from the conversion data corresponding to the candidate recommended content. For example, the preset filtering conditions may include the conversion consumption value reaching a preset consumption threshold, the amount of conversion data within a preset time period reaching a threshold, etc. In this way, the second recommended content that matches the attribute value of the first recommended content under the target attribute category can be effectively obtained.

[0084] In one embodiment, obtaining conversion data corresponding to candidate recommended content includes: determining candidate recommended content that matches the attribute value of the first recommended content under the target attribute category; obtaining conversion data of the candidate recommended content over multiple time periods; the amount of conversion data in each time period is negatively correlated with the distance of the time period from the current time. Here, a time period refers to a historical period, such as one hour, three hours, or one day. The target attribute category includes at least one. Matching attribute values ​​under the target attribute category specifically means that the attribute values ​​under the target attribute category are the same or similar.

[0085] In the process of acquiring conversion data corresponding to candidate recommended content, the computer device first determines the target attribute category of the content to be recommended, and then determines candidate recommended content that matches the attribute values ​​of the first recommended content under one or more target attribute categories. For example, the target attribute category may include at least one of the following: belonging attribute, cluster attribute, content category attribute, promotion site attribute, etc.

[0086] Among these, the attribution attribute can refer to the target audience of the recommended content. For example, if the recommended content is promotional information, its target audience is the merchant to which the promotional information belongs, such as the advertiser. The cluster attribute can refer to the cluster attribute identifier of the recommended content. For example, the cluster attribute of promotional information could be the corresponding brand identifier. The content category attribute can refer to the category to which the recommended content belongs. For example, if the recommended content is product information, its category could be food, clothing, mobile phones, computers, games, or other product categories. The promotion site attribute can refer to the promotion platform used to promote the recommended content, such as various websites, applications, search engines, etc.

[0087] Then, the computer equipment further acquires the conversion data of the candidate recommended content over multiple time periods according to a time decay coefficient. The time decay coefficient is a parameter used to control the amount of conversion data collected for the candidate recommended content, which gradually decreases over time. For example, taking a one-hour time period as an example, the computer equipment can collect 100% of the conversion data corresponding to the candidate recommended content in the first time period closest to the current time; collect 90% of the conversion data in the second time period closest to the current time; collect 80% of the conversion data in the third time period closest to the current time, and so on.

[0088] The time decay coefficient ensures that the amount of conversion data within each time period is negatively correlated with the distance of that time period from the current time. In other words, when collecting conversion data, the closer to the current time, the more conversion data is collected; the further back in time, the less conversion data is collected. This ensures that the collected conversion data is closest to the current time, allowing for a more accurate analysis of conversion deviations corresponding to the second recommended content that best reflects the current situation.

[0089] For example, taking the first recommended content as advertising information, the target attribute categories of the advertising information can include the advertiser, advertising brand, and advertising category. The candidate recommended content obtained by the computer device can be advertising information that shares the same advertiser, advertising brand, and advertising category as the first recommended content. This allows the acquisition of advertising information with the same attribute values ​​as the first recommended content across multiple attribute category dimensions. Specifically, advertising information sharing the same advertiser can be all advertising information under the same advertiser as the first recommended content within the same advertising site set; advertising information sharing the same advertising brand can be all advertising information with the same advertising brand as the first recommended content within the same advertising site set; and advertising information sharing the same advertising category can be all advertising information with the same advertising category as the first recommended content within the same advertising site set.

[0090] For example, computer devices can collect conversion data for recommended content categorized by advertiser, specifically conversion data for recommended content from the same advertiser as the first recommended content. Specifically, conversion data for advertisers within historical time periods can be calculated separately for each site and whether the ad is new or old, such as the conversion count for ads on the current day and in the most recent hour, as well as the sum of estimated conversions corresponding to total valid clicks.

[0091] For example, conversion data for new and old ads can be separately calculated for different site sets 1 and 2. For instance, conversion data for the entire day and the most recent hour can be calculated based on four combined dimensions: [site set 1, new ad], [site set 1, old ad], [site set 2, new ad], and [site set 2, old ad]. When calculating the entire day's data, a relatively straightforward time decay strategy can be used for data collection. The formula can be as follows:

[0092]

[0093]

[0094] Where Conv_ad_day is the sum of all historical conversion data for the current day; I is the current time, which can be in hourly increments; λ is the time decay coefficient, which can be 0.05. conv_ad_hour i PCVR_ad_day is the sum of the total historical conversion data obtained at time i. PCVR_ad_hour is the sum of the total estimated conversion data for the current day. i It is the total estimated transformation data obtained from the statistics at time i.

[0095] In a specific implementation, if sufficient data is available within the current hour, the converted data from the current hour is used; otherwise, data from the entire day is used. This can be expressed using the following formula:

[0096]

[0097]

[0098] Where Conv_ad represents the historical conversion data corresponding to the ads obtained through statistics; conv_ad_hour i It is the total historical conversion data obtained at time i; Conv_ad_day is the sum of the total historical conversion data for the current day; PCVR_ad_hour i PCVR_ad_day is the total estimated conversion data obtained at time i; PCVR_ad_day is the sum of the total estimated conversion data for the day; PCVR_ad is the estimated conversion data corresponding to the advertisement obtained from the statistics.

[0099] In this embodiment, by obtaining the conversion data corresponding to the candidate recommended content according to the time decay coefficient, it is possible to obtain conversion data that is closer to the current time and more consistent with the current situation of the candidate recommended content. Therefore, by utilizing the historical conversion data and estimated conversion data of the second recommended content selected from the candidate content, more accurate calibration information that conforms to the current actual situation can be calculated.

[0100] In one embodiment, such as Figure 5 As shown, another method for processing recommended content is provided, which includes the following steps:

[0101] Step S502: Perform conversion prediction on the first recommended content to obtain the initial predicted conversion data of the first recommended content.

[0102] Step S504: Determine candidate recommended content that matches the attribute value of the first recommended content under the target attribute category; obtain the conversion data of the candidate recommended content over multiple time periods.

[0103] Step S506: When there are multiple target attribute categories, determine the current target attribute category according to the priority order of each target attribute category.

[0104] Step S508: If the candidate recommended content under the current target attribute category does not meet the content extraction conditions, then select the next target attribute category as the current target attribute category.

[0105] Step S510: Until the candidate recommended content under the current target attribute category meets the content extraction conditions, select the second recommended content from the candidate recommended content under the current target attribute category.

[0106] Step S512: Extract the historical conversion data and estimated conversion data corresponding to the second recommended content from the conversion data corresponding to the candidate recommended content.

[0107] Step S514: Based on the difference between historical conversion data and estimated conversion data, determine the calibration information corresponding to the initial estimated conversion data.

[0108] Step S516: Adjust the initial estimated conversion data based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0109] It can be understood that the priority of the target attribute category can be preset priority attribute information for each target attribute category. There is a priority order among multiple target attribute categories.

[0110] Specifically, the target attribute categories can include multiple categories. When there are multiple target attribute categories, the computer device can obtain recommended content that matches the attribute value of the first recommended content under each target attribute category according to the multiple target attribute categories corresponding to the first recommended content, obtain candidate recommended content, and obtain the conversion data corresponding to the candidate recommended content.

[0111] The computer device further extracts second recommended content that meets the content extraction conditions from the candidate recommended content. Specifically, the computer device determines the current target attribute category in order of priority corresponding to each target attribute category. The computer device then determines whether the candidate recommended content under the current target attribute category needs to be extracted based on the preset content extraction conditions. These content extraction conditions may include at least one of the following: the amount of conversion data of the candidate recommended content reaches a preset threshold, and the resource consumption value corresponding to the candidate recommended content reaches a resource consumption threshold. The resource consumption value can be the resources required to recommend the content. For example, taking advertising information as an example, the resource consumption value of the advertising information can be the corresponding advertising bid.

[0112] If the candidate recommended content under the current target attribute category does not meet the content extraction criteria, the computer device selects the next target attribute category as the current target attribute category. This process continues until the candidate recommended content under the current target attribute category meets the content extraction criteria. Then, a second recommended content is selected from the candidate recommended content under the current target attribute category; that is, the recommended content that meets the content extraction criteria is selected as the second recommended content. The computer device then obtains the historical conversion data and estimated conversion data corresponding to the second recommended content from the conversion data.

[0113] For example, taking the first recommended content as advertising information, the target attribute category of the advertising information can include the advertiser, advertising brand, and advertising category. The computer device determines the recommended content that is the same as the advertiser, advertising brand, and advertising category of the first recommended content from the target attribute category dimensions corresponding to the advertiser, advertising brand, and advertising category, and uses it as candidate recommended content, and obtains the conversion data corresponding to the candidate recommended content.

[0114] For example, from the advertiser perspective, conversion data can be collected for candidate recommended content from the same advertiser as the first recommended content, based on different site set attributes and whether the ad is new or old, within a historical time period. This includes the total number of conversions for the day and the most recent hour, as well as the sum of the estimated conversion data corresponding to the total valid clicks. From the advertiser brand perspective, conversion data can be collected for candidate recommended content from the same advertiser brand as the first recommended content, within a historical time period. This includes the sum of the total conversions and the sum of the estimated conversion data corresponding to the total valid clicks. Finally, from the advertiser product category perspective, conversion data can be collected for candidate recommended content from the same advertiser product category as the first recommended content, within a historical time period, including the total conversions and the sum of the estimated conversion data corresponding to the total valid clicks.

[0115] This allows us to collect historical aggregated data of candidate recommendation content pairs under multiple combined dimensions, namely, conversion data corresponding to candidate recommendation content under the target attribute category dimension of advertiser, advertising brand, and advertising category.

[0116] The computer device then selects the second recommended content from the candidate recommended content under multiple target attribute categories according to the priority order corresponding to each target attribute category and the content extraction conditions. For example, the priority order of the target attribute categories corresponding to advertiser, advertising brand, and advertising category is advertiser, advertising category, advertising brand.

[0117] The content extraction criteria can include whether the conversion data volume of candidate recommended content meets a data volume threshold, and whether the actual resource consumption value of candidate recommended content reaches a preset multiple of the target resource consumption value. Specifically, the validity of the conversion data of each candidate recommended content can be determined based on the content extraction criteria. Then, according to the priority order corresponding to each target attribute category and the content extraction criteria, the conversion data corresponding to the second recommended content with the highest priority and validity is extracted from the candidate recommended content.

[0118] In another embodiment, after acquiring the conversion data corresponding to the candidate recommended content, the computer device can further filter out recommended content that meets the content extraction conditions under one or more target attribute categories from the candidate recommended content as second recommended content. Further, the computer device can also filter out recommended content that meets the content extraction conditions under one or more target attribute categories from the candidate recommended content according to the weight of each target attribute category as second recommended content.

[0119] In another embodiment, the computer device may first determine candidate recommended content according to the priority order corresponding to the target attribute category and preset filtering conditions. The computer device further extracts recommended content that meets the content extraction conditions from the candidate recommended content as second recommended content. The computer device then obtains the historical conversion data and estimated conversion data corresponding to the second recommended content.

[0120] In this embodiment, by extracting historical conversion data and estimated conversion data corresponding to the second recommended content that meets the content extraction conditions from the candidate recommended content according to the priority order corresponding to the target attribute category, it is possible to filter out historical conversion data and estimated conversion data corresponding to recommended content that is more relevant or more important to the first recommended content.

[0121] In one embodiment, determining calibration information corresponding to the initial estimated conversion data based on the difference between historical conversion data and estimated conversion data includes: when the first recommended content is in the initial exposure stage, obtaining an amplification factor corresponding to the initial estimated conversion data; and determining calibration information corresponding to the initial estimated conversion data based on the difference between historical conversion data and estimated conversion data and the amplification factor.

[0122] The amplification factor, also known as the gain factor, refers to the information used to amplify calibration information. For example, it can be used to amplify the numerical values ​​in calibration information by a factor corresponding to the amplification factor. The amplification factor is also called gain, which is the ratio of the change in output to the change in input. It reflects steady-state characteristics; the larger the amplification factor, the greater the influence of the input variable on the output variable.

[0123] The initial exposure phase can be understood as the stage where the historical exposure of the recommended content has not yet reached the preset exposure threshold. Here, exposure refers to the number of times the recommended content is displayed, and clicks refer to the number of times the recommended content is clicked; the ratio of clicks to exposures is the click-through rate (CTR).

[0124] Specifically, when the first recommended content is in the initial exposure stage, the computer device performs a conversion prediction on the first recommended content to obtain initial predicted conversion data. The computer device then acquires the historical conversion data and predicted conversion data corresponding to the second recommended content, and subsequently obtains the scaling factor corresponding to the initial predicted conversion data. Then, based on the difference between the historical conversion data and the predicted conversion data, as well as the scaling factor, calibration information corresponding to the initial predicted conversion data is determined.

[0125] In another embodiment, the computer device may further adjust the calibration information using an amplification factor after determining the calibration information corresponding to the initial estimated conversion data based on the difference between the historical conversion data and the estimated conversion data corresponding to the second recommended content. The adjusted calibration factor is then used as the final calibration information corresponding to the initial estimated conversion data.

[0126] In one embodiment, obtaining the expansion coefficient corresponding to the initial estimated conversion data includes: determining the target resource consumption range corresponding to the first recommended content from a preset resource consumption range; and determining the expansion coefficient corresponding to the target resource consumption range.

[0127] The preset resource consumption range refers to multiple ranges of resource consumption values ​​pre-defined based on actual resource consumption values. Specifically, it can be pre-defined based on the difference between the actual resource consumption value and the target resource consumption value of the first recommended content. The target resource consumption range refers to one of the preset resource consumption ranges where the resource consumption value of the recommended content falls. It can be understood that the first recommended content includes both the corresponding actual resource consumption value and the target resource consumption value.

[0128] The actual resource consumption value refers to the actual single-unit resource consumption of the first recommended content over a historical period, that is, the resource value consumed by each first recommended content. The target resource consumption value refers to the pre-set expected single-unit resource consumption value for the first recommended content, that is, the pre-set expected resource consumption value for each first recommended content.

[0129] Specifically, when the first recommended content is in the initial exposure stage, the computer device acquires the actual resource consumption value of the first recommended content, and then determines the target resource consumption range corresponding to the actual resource consumption value of the first recommended content within a preset resource consumption range based on the difference between the actual resource consumption value and the target resource consumption value. The computer device then determines an expansion coefficient corresponding to the target resource consumption range based on the target resource consumption range to which the first recommended content actually belongs.

[0130] In this embodiment, by determining the target resource consumption range of the first recommended content based on the difference between the actual resource consumption value and the target resource consumption value, the target resource consumption range of the first recommended content can be effectively determined. This allows for the acquisition of historical conversion data and estimated conversion data of the second recommended content within the target resource consumption range of the first recommended content. Consequently, more accurate calibration information can be calculated based on the expansion coefficient corresponding to the target resource consumption range.

[0131] In one embodiment, determining the expansion coefficient corresponding to the target resource consumption range includes: obtaining the deviation between the current actual total resource consumption and the current target total resource consumption; the current actual total resource consumption is the total amount of resources currently actually consumed by the first recommended content; the current target total resource consumption is the total amount of target resource consumption currently corresponding to the first recommended content; and determining the expansion coefficient corresponding to the target resource consumption range based on the deviation.

[0132] Among them, the current actual total resource consumption refers to the total amount of resources currently actually consumed by the first recommended content; the current target total resource consumption refers to the total amount of target resources currently consumed by the first recommended content.

[0133] After determining the target resource consumption range to which the first recommended content actually belongs, the computer equipment also obtains the current actual resource consumption and the current target resource consumption corresponding to the first recommended content, and determines the deviation between the current actual resource consumption and the current target resource consumption. Then, based on the deviation between the current actual resource consumption and the current target resource consumption, it determines the expansion coefficient of the first recommended content corresponding to the target resource consumption range.

[0134] For example, the resource consumption value of the first recommended content can be divided into multiple preset resource consumption intervals. For instance, it could be divided into three preset resource consumption intervals. The first resource consumption interval can be the stage where the actual resource consumption value of the first recommended content is lower than a first preset multiple of the target resource consumption value; the second resource consumption interval can be the stage where the actual resource consumption value of the first recommended content is between the first preset multiple of the target resource consumption value and the second preset multiple of the target resource consumption value; and the third resource consumption interval can be the stage where the actual resource consumption value of the first recommended content is greater than a third preset multiple of the target resource consumption value.

[0135] Taking the primary recommended content as an advertisement as an example, the actual resource consumption value of the advertisement is the actual resources consumed, while the target resource consumption value can be the target bid. The resource consumption of the advertisement can be divided into three stages: the first stage is when the actual resource consumption value is less than 8 times the target resource consumption value; the second stage is when the actual resource consumption value is between 8 and 25 times the target resource consumption value; and the third stage is when the actual resource consumption value is greater than 25 times the target resource consumption value.

[0136] When the target resource consumption range corresponding to the first recommended content is within the first resource consumption range, the computer device can determine the expansion factor corresponding to the first resource consumption range based on a preset value, such as 1. It is understood that when the target resource consumption range of the first recommended content is within the first resource consumption range, it indicates that the resource consumption and total data volume at this stage are relatively small, making it easily susceptible to interference and leading to larger errors in the calibration information, ultimately exceeding the resource consumption cost. Therefore, setting the expansion factor to 1 is equivalent to not expanding the calibration information.

[0137] When the target resource consumption interval corresponding to the first recommended content is the second resource consumption interval, the expansion coefficient corresponding to the second resource consumption interval is calculated based on the deviation between the current actual total resource consumption and the current target total resource consumption.

[0138] In this embodiment, when the first recommended content is in the initial exposure stage, the resource consumption range is divided according to the difference between the actual resource consumption value and the target resource consumption value of the first recommended content. Then, the expansion coefficient is calculated based on the deviation between the current actual total resource consumption and the current target total resource consumption of the first recommended content. This can effectively obtain a more accurate expansion coefficient that matches the exposure situation of the first recommended content itself, thereby obtaining more accurate calibration information for the initial estimated conversion data of the first recommended content.

[0139] In one embodiment, determining calibration information corresponding to the initial estimated conversion data based on the difference between historical conversion data and estimated conversion data includes: obtaining a smoothing coefficient when the first recommended content is in a fully exposed stage; and determining the calibration information corresponding to the initial estimated conversion data based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data. The smoothed historical conversion data is obtained by smoothing the historical conversion data corresponding to the second recommended content according to the smoothing coefficient; the smoothed estimated conversion data is obtained by smoothing the estimated conversion data corresponding to the second recommended content according to the smoothing coefficient.

[0140] The "full exposure stage" refers to the stage where the historical exposure of the recommended content reaches a preset exposure threshold. The smoothing coefficient, a factor related to exponential smoothing, determines the smoothing level and the response speed to differences between predicted and actual results. A larger smoothing coefficient means a more rapid decrease in the influence of future actual values ​​on the current data to be smoothed; a smaller smoothing coefficient means a slower decrease in the influence of future actual values ​​on the current data to be smoothed. In this embodiment, the smoothing coefficient refers to the smoothing coefficient applied to historical conversion data.

[0141] The smoothing coefficient is the smoothing coefficient corresponding to the first recommended content. It can be determined based on the historical conversion data corresponding to the first recommended content and is used to smooth the historical conversion data and estimated conversion data corresponding to the second recommended content.

[0142] Specifically, when the first recommended content is in the full exposure stage, the computer device performs a conversion prediction on the first recommended content, obtains the initial estimated conversion data of the first recommended content, and after obtaining the historical conversion data and estimated conversion data corresponding to the second recommended content, the computer device also obtains the historical conversion data corresponding to the first recommended content.

[0143] The historical conversion data corresponding to the first recommended content may include the total number of historical conversions and exposure duration for the first recommended content within a historical time period. The computer then determines the smoothing coefficient corresponding to the first recommended content based on the total number of historical conversions and exposure duration.

[0144] The computer equipment then smooths the historical conversion data and estimated conversion data corresponding to the second recommended content based on the smoothing coefficient corresponding to the first recommended content, to obtain smoothed historical conversion data and smoothed estimated conversion data.

[0145] The computer device then determines calibration information corresponding to the first initial estimated conversion data based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data. Thus, by using a smoothing coefficient determined based on the historical conversion data of the first recommended content, and by addressing the difference between the smoothed historical conversion data and the estimated conversion data corresponding to the second recommended content, more accurate calibration information for the initial estimated conversion data of the first recommended content can be obtained.

[0146] In one embodiment, the above-mentioned method for processing recommended content further includes: determining resource consumption deviation information based on the resource consumption level corresponding to the first recommended content, and the historical conversion data and estimated conversion data of the second recommended content; adjusting the historical conversion data of the second recommended content based on the resource consumption deviation information to obtain adjusted historical conversion data; and smoothing the adjusted historical conversion data corresponding to the second recommended content according to a smoothing coefficient to obtain smoothed historical conversion data.

[0147] The resource consumption level refers to the resource consumption level corresponding to the actual resource consumption value of the recommended content within the preset resource consumption levels. The target resource consumption level represents the resource consumption level corresponding to the current actual resource consumption value of the first recommended content within the preset resource consumption levels, which can be determined based on the difference between the actual resource consumption value and the target resource consumption value of the recommended content.

[0148] The computer device determines the corresponding smoothing coefficient based on the historical conversion data corresponding to the first recommended content, and then obtains the actual resource consumption value and the target resource consumption value corresponding to the first recommended content. Then, based on the difference between the actual resource consumption value and the target resource consumption value of the first recommended content, it determines the current target resource consumption level corresponding to the actual resource consumption value of the first recommended content within a preset resource consumption level.

[0149] Similarly, the computer device can also classify the second recommended content into preset resource consumption levels in the same way. The computer device then extracts the historical conversion data and estimated conversion data under the target resource consumption level of the first recommended content from the total historical conversion data and estimated conversion data of the second recommended content, and then calculates the deviation between the historical conversion data and estimated conversion data under the target resource consumption level and the total historical conversion data and estimated conversion data of the second recommended content, thereby obtaining resource consumption deviation information.

[0150] The computer equipment then uses resource consumption deviation information to adjust the historical conversion data of the second recommended content, thereby obtaining adjusted historical conversion data. Next, the computer equipment uses the smoothing coefficient corresponding to the first recommended content to smooth the adjusted historical conversion data corresponding to the second recommended content, obtaining smoothed historical conversion data.

[0151] In this embodiment, the historical conversion data corresponding to the second recommended content is adjusted based on the target resource consumption level corresponding to the first recommended content, and the adjusted historical conversion data of the second recommended content is smoothed using the smoothing coefficient corresponding to the first recommended content. Based on the difference between the smoothed historical conversion data and the estimated conversion data corresponding to the second recommended content, calibration information that is more consistent with the current resource consumption level of the first recommended content can be obtained.

[0152] In one embodiment, determining resource consumption deviation information based on the target resource consumption level corresponding to the first recommended content and the historical conversion data and estimated conversion data of the second recommended content includes: extracting historical conversion data and estimated conversion data under the target resource consumption level from the historical conversion data and estimated conversion data of the second recommended content, based on the target resource consumption level corresponding to the first recommended content; and determining resource consumption deviation information based on the difference between the historical conversion data and estimated conversion data under the target resource consumption level and the historical conversion data and estimated conversion data of the second recommended content.

[0153] Specifically, the computer device acquires the actual and target resource consumption values ​​of the second recommended content within a historical time period. Then, based on the difference between the actual and target resource consumption values, it determines the various resource consumption levels corresponding to the actual resource consumption value of the second recommended content within a preset resource consumption level. Thus, according to the resource consumption level, the total historical conversion data and estimated conversion data corresponding to the second recommended content can be divided into historical conversion data and estimated conversion data under each resource consumption level.

[0154] The computer equipment then determines the resource consumption level corresponding to the target resource consumption level of the first recommended content from among the various resource consumption levels of the second recommended content. Then, from the total historical conversion data and estimated conversion data corresponding to the second recommended content, it extracts the historical conversion data and estimated conversion data under the resource consumption level corresponding to the target resource consumption level of the first recommended content.

[0155] The computer equipment then determines resource consumption deviation information based on the difference between the historical and estimated conversion data of the second recommended content at the target resource consumption level and the total historical and estimated conversion data of the second recommended content. This allows for the generation of resource consumption deviation information corresponding to the second recommended content that matches the current actual resource consumption value of the first recommended content.

[0156] In one embodiment, performing conversion prediction on the first recommended content to obtain initial predicted conversion data for the first recommended content includes: obtaining content attribute information and recommendation object information corresponding to the first recommended content; extracting content features from the content attribute information and object features from the recommendation object information using a trained conversion prediction model; and performing conversion prediction on the first recommended content based on the content features and object features to obtain initial predicted conversion data for the first recommended content.

[0157] It can be understood that a trained conversion prediction model refers to a machine learning model that has been pre-trained and has the ability to predict conversion data for recommended content. Specifically, it can be trained based on a large amount of historical recommended content and corresponding historical conversion and access data.

[0158] After obtaining the first recommended content, the computer device first performs conversion prediction processing on the first recommended content. Specifically, the computer device inputs the content attribute information and recommendation object information corresponding to the first recommended content into the trained conversion prediction model.

[0159] The computer equipment then uses a conversion prediction model to estimate the initial conversion data of the first recommended content based on content attribute information and recommendation object information. Specifically, the conversion prediction model extracts corresponding content features from the content attribute information and corresponding object features from the recommendation object information. It then performs conversion prediction processing on the first recommended content based on these content and object features, and finally outputs the initial conversion data corresponding to the first recommended content.

[0160] In one embodiment, the conversion prediction model is obtained through a model training step, such as... Figure 6 As shown, the steps for model training include:

[0161] Step S602: Obtain sample content and training labels; the sample content includes historical recommended content, and the training labels are determined based on the historical conversion data and historical access data corresponding to the historical recommended content.

[0162] Step S604: Extract sample content features and sample object features corresponding to the sample content using the conversion prediction model to be trained in this round.

[0163] Step S606: Based on the characteristics of the sample content and the characteristics of the sample object, perform a conversion prediction on the sample content to obtain the sample conversion prediction data.

[0164] Step S608: Based on the difference between the sample predicted conversion data and the training labels, adjust the parameters of the conversion prediction model, take the next round as the current round and continue training until the training conditions are met and then stop training.

[0165] The sample content can be historical data corresponding to multiple historical recommended content items within a historical time period. Historical data includes historical content attribute information and historical recommendation object information for each historical recommended content item, as well as corresponding historical conversion data and historical access data. Historical conversion data refers to the actual conversion data for each historical recommended content item within a historical time period; historical access data refers to the actual number of times each historical recommended content item was accessed by users within a historical time period.

[0166] Computer devices can collect historical data corresponding to historical recommended content within a historical time period, and use the historical content attribute information and historical recommended object information corresponding to the historical recommended content in the historical data as sample content for training the conversion prediction model.

[0167] After acquiring historical conversion and access data for multiple historical recommended content items, the computer device can determine the actual historical conversion rate of each historical recommended content item within a historical time period based on the historical conversion and access data for each item. Then, the historical conversion rate corresponding to each historical recommended content item is used as the training label for training the conversion prediction model.

[0168] Specifically, the computer device inputs each historical recommendation, including its corresponding historical content attribute information and historical recommendation object information, into the conversion prediction model to be trained. The conversion prediction model extracts corresponding sample content features based on the sample content attribute information of each historical recommendation and corresponding sample object features based on the sample recommendation object information. Then, the conversion prediction model predicts the sample conversion data corresponding to each historical recommendation based on the sample content features and sample object features.

[0169] In one embodiment, the conversion prediction model includes a first extraction network, a second extraction network, and a prediction network layer. The first and second extraction networks can specifically be deep neural network models, and the prediction network layer can specifically be a linear model. The first extraction network of the conversion prediction model extracts features from the sample content attribute information to obtain corresponding sample content features; the second extraction network of the conversion prediction model extracts features from the sample recommendation object information to obtain corresponding sample object features. Then, based on the sample content features and sample object features, the prediction network layer of the conversion prediction model predicts the sample conversion data corresponding to each historical recommendation content.

[0170] Then, the computer equipment determines the difference between the sample estimated conversion data and the training labels based on the sample estimated conversion data and historical conversion rate corresponding to each historical recommended content. Subsequently, based on the difference between the sample estimated conversion data and the training labels, the parameters of the conversion prediction model are adjusted, and the next round is used as the current round for continued training until the training conditions are met, at which point training stops.

[0171] The difference between the sample click-through rate and the efficiency label can be measured using a loss function, such as the Mean Absolute Value Loss (MAE), the smoothed mean absolute error (Huber loss), or the cross-entropy loss function. Training conditions are the conditions for ending model training. Training stopping conditions can be reaching a preset number of iterations, or the predicted performance index of the conversion prediction model after parameter adjustments reaching a preset index.

[0172] In this embodiment, the computer device trains the conversion prediction model to be trained using historical data corresponding to historical recommended content. It then gradually adjusts the parameters of the conversion prediction model by utilizing the difference between the actual historical conversion rate of the historical recommended content and the sample predicted conversion data. This parameter adjustment process enables the conversion prediction model to capture the implicit relationship between sample content features, sample object features, and the predicted conversion data of the recommended content, and gradually reduces the difference between the sample predicted conversion data and the actual historical conversion rate, resulting in higher accuracy in conversion prediction.

[0173] In a specific embodiment, such as Figure 7 As shown, a specific method for processing recommended content is provided, which includes the following steps:

[0174] Step S702: Obtain the content attribute information and recommendation object information corresponding to the first recommended content.

[0175] Step S704: Extract content features of content attribute information and object features of recommendation object information using the trained conversion prediction model.

[0176] Step S706: Based on content features and object features, perform conversion prediction on the first recommended content to obtain the initial predicted conversion data of the first recommended content.

[0177] Step S708: Determine candidate recommended content that matches the attribute value of the first recommended content under the target attribute category, and obtain the conversion data of the candidate recommended content in multiple time periods; extract the historical conversion data and estimated conversion data corresponding to the second recommended content from the conversion data.

[0178] Step S710: When the first recommended content is in the initial exposure stage, determine the target resource consumption range corresponding to the first recommended content from the preset resource consumption range.

[0179] Step S712: Obtain the deviation between the current actual total resource consumption and the current target total resource consumption.

[0180] Step S714: Determine the expansion coefficient corresponding to the target resource consumption range based on the deviation.

[0181] Step S716: Based on the difference between historical conversion data and estimated conversion data, as well as the amplification factor, determine the calibration information corresponding to the initial estimated conversion data.

[0182] Step S718: When the first recommended content is in the full exposure stage, obtain the smoothing coefficient corresponding to the first recommended content; extract the historical conversion data and estimated conversion data of the second recommended content under the target resource consumption level corresponding to the first recommended content from the historical conversion data and estimated conversion data of the second recommended content.

[0183] Step S720: Based on the differences between the historical and estimated conversion data under the target resource consumption level and the historical and estimated conversion data of the second recommended content, determine the resource consumption deviation information.

[0184] Step S722: Based on the resource consumption deviation information, adjust the historical conversion data of the second recommended content, and smooth the adjusted historical conversion data according to the smoothing coefficient to obtain smoothed historical conversion data.

[0185] Step S724: Based on the difference between the smoothed historical conversion data and the smoothed predicted conversion data based on the smoothing coefficient, determine the calibration information corresponding to the initial predicted conversion data.

[0186] Step S726: Adjust the initial estimated conversion data based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0187] In this embodiment, the computer device performs conversion prediction on the first recommended content. After obtaining the initial predicted conversion data of the first recommended content, it acquires the conversion data corresponding to candidate recommended content that matches the attribute value of the first recommended content under the target attribute category. It then extracts the historical conversion data and predicted conversion data corresponding to the second recommended content that meets the content advance conditions. Furthermore, based on the different exposure stages of the first recommended content, different methods are used to determine calibration information corresponding to the initial predicted conversion data based on the difference between the historical conversion data and the predicted conversion data. This effectively utilizes the historical data of the second recommended content under the same target attribute category to calibrate and correct the initial predicted conversion data of the first recommended content, reducing the error caused by directly predicting the conversion of the recommended content based on its own data. This results in a more accurate conversion prediction result, thereby effectively improving the accuracy of conversion prediction for recommended content.

[0188] In one example embodiment, taking recommended content as advertising information, the first recommended content is the first advertising information. The computer device obtains the content attribute information and recommendation object information corresponding to the first advertising information, and uses a trained conversion prediction model to predict the initial predicted conversion data of the first advertising information based on the content attribute information and recommendation object information.

[0189] The computer device then determines multiple target attribute categories corresponding to the first advertising information. These could be target attribute categories corresponding to the advertiser, advertising brand, and advertising category. It then retrieves candidate advertising information that matches the advertiser, advertising brand, and advertising category of the first advertising information, respectively, within the dimensions of the target attribute categories corresponding to the advertiser, advertising brand, and advertising category. Specifically, these can be candidate advertising information that shares the same advertiser, advertising brand, and advertising category as the first advertising information.

[0190] Specifically, candidate recommended content with the same advertiser, brand, and category as the first advertisement is collected from the dimensions of advertiser, brand, and category. Historical conversion data and estimated conversion data within the historical time period are collected, such as the total number of advertisement conversions on the same day and in the most recent hour, and the sum of the estimated conversion data corresponding to the total effective clicks.

[0191] The computer device then selects the highest priority advertisement information that meets the content extraction conditions from the candidate advertisement information under multiple target attribute categories according to the priority order corresponding to each target attribute category and the content extraction conditions, as the second advertisement information, and obtains the historical conversion data and estimated conversion data corresponding to the second advertisement information.

[0192] The computer equipment can then determine calibration information corresponding to the initial estimated conversion data of the first advertising information based on the difference between the historical conversion data and the estimated conversion data corresponding to the second advertising information. Specifically, the calibration information can be a calibration coefficient. For example, the calibration information can be calculated using the following formula:

[0193] cali_rate=Conv_valid / PCVR_valid;

[0194] Where cali_rate represents the calibration coefficient corresponding to the initial estimated conversion data of the first advertising information, Conv_valid represents the historical conversion data corresponding to the second advertising information, and PCVR_valid represents the estimated conversion data corresponding to the second advertising information.

[0195] When the first advertisement is in its initial exposure phase, the expansion coefficient corresponding to it is obtained. The expansion coefficient corresponding to the target resource consumption range can be determined based on the deviation between the current actual resource consumption and the current target resource consumption of the first advertisement. The actual resource consumption value of the advertisement is the actual resources consumed, and the target resource consumption value can be the target bid of the advertisement.

[0196] Specifically, the resource consumption of advertising information is divided into three stages. For example, the first resource consumption range is when the actual resource consumption value of the advertising information is less than 8 times the target resource consumption value; the second resource consumption range is when the actual resource consumption value of the advertising information is between 8 times and 25 times the target resource consumption value; and the third resource consumption range is when the actual resource consumption value of the advertising information is greater than 25 times the target resource consumption value.

[0197] If the target resource consumption range corresponding to the first advertising information is within the first resource consumption range, an expansion coefficient corresponding to the first resource consumption range can be determined based on a preset value, such as 1. It is understood that when the target resource consumption range of the first advertising information is within the first resource consumption range, it indicates that the resource consumption and overall data volume at that stage are relatively small, making it easily susceptible to interference and leading to larger errors in the calibration information, ultimately exceeding the resource consumption cost. Therefore, setting the expansion coefficient to 1 is equivalent to not expanding the calibration information.

[0198] If the target resource consumption range corresponding to the first advertising information is the second resource consumption range, the expansion coefficient can be obtained using the following formula:

[0199]

[0200] If the target resource consumption range corresponding to the first advertising information is the third resource consumption range, the expansion coefficient can be obtained using the following formula:

[0201]

[0202] Wherein, *coef* represents the expansion coefficient corresponding to the initial estimated conversion data of the first advertising information; *t_cpa* refers to the target resource consumption value corresponding to the first advertising information, such as the target bid of the advertising information; and *cpa_bias* refers to the current target bid deviation corresponding to the first advertising information, such as the current target bid deviation of the advertising information. The current target bid deviation can be obtained using the following formula:

[0203] cpa_bias=cost / (conv_num*t_cpa)–1;

[0204] Where cost is the total amount of resources actually consumed by the first advertising information, such as the current consumption of the advertising information; conv_num is the total amount of conversion data corresponding to the first advertising information, such as the current number of conversions of the advertising information; conv_num*t_cpa represents the total amount of target resources consumed by the first advertising information.

[0205] When the first advertisement is in its full exposure phase, the computer device acquires the smoothing coefficient corresponding to the first advertisement. Specifically, it can acquire historical conversion data of the first advertisement within a historical time period, including the exposure duration of the first advertisement within that historical time period. Based on the historical conversion data and the corresponding exposure duration, the smoothing coefficient corresponding to the first advertisement is determined. For example, the smoothing coefficient can be calculated using the following formula:

[0206]

[0207] Wherein, smooth_base represents the smoothing coefficient corresponding to the first ad message, ceil function represents rounding up, sum_conv_yesterday represents the sum of historical conversion data of the first ad message in the previous day, and H represents the exposure duration of the first ad message in the previous day.

[0208] The computer equipment then uses a smoothing coefficient to smooth the historical conversion data and estimated conversion data corresponding to the second advertisement information, obtaining smoothed historical conversion data and smoothed estimated conversion data. Then, based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data, a calibration coefficient corresponding to the initial estimated conversion data of the first advertisement information is determined. For example, the calibration coefficient can be calculated using the following formula:

[0209] cali_rate=(conv_num+smooth_base) / (sum_pcvr+smooth_base);

[0210] Wherein, cali_rate represents the calibration coefficient corresponding to the initial estimated conversion data of the first advertising information, smooth_base represents the smoothing coefficient corresponding to the first advertising information, conv_num represents the historical conversion data corresponding to the second advertising information, where conv_num can be equivalent to conv_valid mentioned above; pcvr_num represents the estimated conversion data corresponding to the second advertising information, where pcvr_num can be equivalent to PCVR_valid mentioned above.

[0211] In a specific test embodiment, the initial estimated conversion data of the first advertising information is adjusted using the calibration coefficient determined in the above manner to obtain the corresponding estimated conversion result. The estimated deviation data between this result and the actual conversion data of the first advertising information can be as follows:

[0212] As shown in Table 1 below, Table 1 displays the average prediction deviation of the first advertising information across four time periods throughout the day, under the dimension of the target attribute category of the advertising identifier, after calibrating the initial estimated conversion data of the first advertising information using calibration coefficients. Here, bias represents the prediction deviation data corresponding to the first advertising information, and abs(bias-1) represents the average prediction deviation obtained by weighting the prediction deviation data corresponding to the first advertising information.

[0213] Time period abs(bias-1) consumption weighted 0～6 0.301428 6～12 0.198272 12～18 0.151886 18～24 0.118574

[0214] Table 1

[0215] As shown in Table 2 below, Table 2 displays the distribution of the estimated deviation data for the first advertising information across four time periods throughout the day, under the dimension of the target attribute category of the advertising identifier, for each resource consumption interval. Here, ">1.5" indicates the resource consumption interval corresponding to the actual resource consumption value of the first advertising information being greater than 1.5 times the target resource consumption value, and so on. "<=1.5&>1.2", "<=1.2&>1.1", "<=1.1&>1", etc., all represent the corresponding resource consumption intervals.

[0216]

[0217]

[0218] Table 2

[0219] As shown in Table 3 below, after calibrating the initial estimated conversion data of the first advertising information using calibration coefficients, the average estimated deviation of the first advertising information in the four time periods throughout the day under the dimension of the advertiser's target attribute category.

[0220] Time period abs(bias-1) consumption weighted 0～6 0.210224 6～12 0.131000 12～18 0.108330 18～24 0.085859

[0221] Table 3

[0222] As shown in Table 4 below, Table 4 displays the distribution of estimated deviation data for the first advertising information across four time periods throughout the day, under the dimension of the advertiser's target attribute category.

[0223]

[0224] Table 4

[0225] Furthermore, the computer device also acquires resource consumption deviation information corresponding to the second advertising information. Specifically, the resource consumption of the advertising information can be divided into multiple resource consumption levels. This can be based on the difference between the actual resource consumption value and the target resource consumption value of the advertising information. For example, as shown in Table 5 below, the resource consumption levels can be divided into the following five tiers:

[0226] Resource Consumption Level Resource consumption value Conversion data at each resource consumption level 1 Cost<=4*t_cpa Conv_ad_1、PCVR_ad_1 2 4*t_cpa<Cost<=10*t_cpa Conv_ad_2、PCVR_ad_2 3 10*t_cpa<Cost<=20*t_cpa Conv_ad_3、PCVR_ad_3 4 20*t_cpa<Cost<=40*t_cpa Conv_ad_4、PCVR_ad_4 5 Cost > 40 * t_cpa Conv_ad_5、PCVR_ad_5

[0227] Table 5

[0228] Wherein, Cost represents the actual resource consumption of the ad message, t_cpa represents the target resource consumption of the ad message, Conv_ad_1 represents the historical conversion data of the ad message at resource consumption level 1, and PCVR_ad_1 represents the estimated conversion data of the ad message at resource consumption level 1. Similarly, Conv_ad_2, Conv_ad_3, Conv_ad_4, Conv_ad_5, etc., represent the historical conversion data at the corresponding resource consumption levels, and PCVR_ad_2, PCVR_ad_3, PCVR_ad_4, PCVR_ad_5, etc., represent the estimated conversion data at the corresponding resource consumption levels.

[0229] Based on the difference between the actual resource consumption value and the target resource consumption value of the first advertising information, the target resource consumption level of the first advertising information is determined. Then, from the total historical conversion data and estimated conversion data of the second advertising information, the historical conversion data and estimated conversion data under the target resource consumption level of the first advertising information are extracted. Then, the deviation between the historical conversion data and estimated conversion data of the second advertising information under the target resource consumption level and the total historical conversion data and estimated conversion data of the second advertising information are calculated, thereby obtaining the resource consumption deviation information.

[0230] For example, the resource consumption deviation information can be calculated using the following formula:

[0231] pcvr_bias_factor=(Conv_valid_p / PCVR_valid_p) / (Conv_valid / PCVR_valid);

[0232] Wherein, pcvr_bias_factor represents the resource consumption deviation information corresponding to the second advertising information, p represents the target resource consumption level of the first advertising information; Conv_valid_p represents the historical conversion data of the second advertising information at resource consumption level p; and PCVR_valid_p represents the estimated conversion data of the second advertising information at resource consumption level p.

[0233] The computer equipment then adjusts the historical conversion data of the second advertisement based on resource consumption deviation information, smoothing the adjusted historical conversion data according to a smoothing coefficient to obtain smoothed historical conversion data. It then smooths the estimated conversion data of the second advertisement according to the smoothing coefficient to obtain smoothed estimated conversion data. Finally, based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data, calibration information corresponding to the initial estimated conversion data is determined. For example, the calibration coefficient can be calculated using the following formula:

[0234]

[0235] Wherein, smooth_base represents the smoothing coefficient corresponding to the first ad information, pcvr_bias_factor represents the resource consumption deviation information corresponding to the second ad information, conv_num represents the historical conversion data corresponding to the second ad information, and pcvr_num represents the estimated conversion data corresponding to the second ad information.

[0236] In a specific test embodiment, the initial estimated conversion data of the first advertising information is adjusted using the calibration coefficient determined in the above manner to obtain the corresponding estimated conversion result. The estimated deviation data between this result and the actual conversion data of the first advertising information can be as follows:

[0237] As shown in Table 6 below, after calibrating the initial estimated conversion data of the first advertising information using the calibration coefficients mentioned above, the average estimated deviation of the first advertising information in the four time periods throughout the day under the dimension of the target attribute category of the advertising identifier.

[0238] Time period abs(bias-1) consumption weighted 0～6 0.303018 6～12 0.184166 12～18 0.141808 18～24 0.109555

[0239] Table 6

[0240] As shown in Table 7 below, Table 7 displays the specific distribution of the estimated deviation data for the first advertising information across four time periods throughout the day, under the dimension of the target attribute category of the advertising identifier.

[0241]

[0242]

[0243] Table 7

[0244] As shown in Table 8 below, after calibrating the initial estimated conversion data of the first advertising information using calibration coefficients, the average estimated deviation of the first advertising information in the four time periods throughout the day under the dimension of the advertiser's target attribute category.

[0245] Time period abs(bias-1) consumption weighted 0～6 0.200226 6～12 0.130225 12～18 0.105439 18～24 0.080237

[0246] Table 8

[0247] As shown in Table 9 below, Table 9 displays the specific estimated deviation data distribution of the first advertising information across four time periods throughout the day, under the dimension of the advertiser's target attribute category.

[0248]

[0249]

[0250] Table 9

[0251] Based on the aforementioned prediction deviation data, the average prediction deviation shows that the predicted conversion data is more accurate across all time periods each day, especially in later time slots where the calibration is more intensive. This is because later time slots require more second-advertisement ads with attribute values ​​matching the first ad's target category to calibrate the single-time prediction results of the first ad. Therefore, later time slots yield richer ad audience information for the second-advertisement ads, resulting in higher overall accuracy of the predicted conversion data.

[0252] Secondly, the distribution of the specific prediction deviation data shows that, for the same time period, the proportion of overestimation in the prediction deviation has been reduced to some extent; and for all time periods, the prediction deviation data can be effectively calibrated. Therefore, combining the historical conversion data and predicted conversion data of the second advertisement with the same attribute values ​​under the target attribute category as the first advertisement to calibrate the initial predicted conversion data of the first advertisement during the full exposure phase can reduce deviation to a certain extent, thus obtaining more accurate conversion prediction results.

[0253] This application also provides an application scenario in which the above-described method for processing recommended content is applied. Specifically, the recommended content includes, but is not limited to, push notifications and promotional information. Push notifications include video content, article content, content relevant to specific user groups, etc. Promotional information can be product information, advertising information, etc.

[0254] The computer device first acquires the first recommended content, which includes corresponding content attribute information and recommendation object information. Based on the content attribute information and recommendation object information, the computer device performs a conversion prediction on the first recommended content, obtaining initial predicted conversion data. Then, the computer device acquires the conversion data corresponding to candidate recommended content that matches the attribute values ​​of the first recommended content under the target attribute category, and extracts the historical conversion data and predicted conversion data corresponding to the second recommended content that meets the content prediction criteria. Furthermore, based on the difference between the historical conversion data and the predicted conversion data, calibration information corresponding to the initial predicted conversion data is determined, and the initial predicted conversion data of the first recommended content is calibrated, adjusted, and corrected using the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0255] Then, when the computer device continues to recommend the first recommended content, it sorts and recommends the first recommended content according to the conversion prediction results. Furthermore, it determines whether to display the content to users based on the conversion prediction results, thereby identifying the target audience for the first recommended content, and then displaying the sorted first recommended content to the target audience.

[0256] If the primary recommended content is promotional information, the computer device can also determine the promotional resource consumption of the primary recommended content based on the conversion forecast and the estimated resource consumption of the primary recommended content, and then promote the primary recommended content based on the promotional resource consumption. For example, it can specifically conduct bidding promotion of the primary recommended content based on the promotional resource consumption to determine the target promotion sites for the primary recommended content, and thus recommend the primary recommended content on the target promotion sites for promotion.

[0257] Furthermore, by combining the conversion data of the second recommended content that matches the attribute value of the first recommended content under the target attribute category, and using calibration information obtained from the difference between the historical conversion data and the estimated conversion data of the second recommended content, the initial estimated conversion data of the first recommended content is adjusted. This effectively reduces the error caused by directly estimating the conversion of recommended content based on the data of the first recommended content itself, thereby obtaining a more accurate conversion prediction result and effectively improving the accuracy of conversion prediction for recommended content.

[0258] It should be understood that, although Figure 3 , 5The steps in the flowchart (Figure 7) are shown sequentially as indicated by the arrows; however, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders. Furthermore, Figure 3 , 5 At least some of the steps in -7 may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0259] In one embodiment, such as Figure 8 As shown, a recommended content processing device 800 is provided. This device can be a software module, a hardware module, or a combination of both as part of a computer device. Specifically, the device includes: a conversion prediction module 802, a data acquisition module 804, a calibration information determination module 806, and a data calibration module 808, wherein:

[0260] The conversion prediction module 802 is used to predict the conversion of the first recommended content and obtain the initial predicted conversion data of the first recommended content.

[0261] The data acquisition module 804 is used to acquire historical conversion data and estimated conversion data corresponding to the second recommended content; the second recommended content is recommended content with the same content attributes as the first recommended content.

[0262] The calibration information determination module 806 is used to determine the calibration information corresponding to the initial estimated conversion data based on the difference between historical conversion data and estimated conversion data.

[0263] The data calibration module 808 is used to adjust the initial estimated conversion data based on calibration information to obtain the conversion prediction result corresponding to the first recommended content.

[0264] In one embodiment, the data acquisition module 802 is further configured to acquire conversion data corresponding to candidate recommended content; candidate recommended content is recommended content that matches the attribute value of the first recommended content under the target attribute category; conversion data includes historical conversion data and estimated conversion data; from the conversion data, historical conversion data and estimated conversion data corresponding to the second recommended content are extracted; the second recommended content is recommended content among the candidate recommended content that meets the preset filtering conditions.

[0265] In one embodiment, the data acquisition module 802 is further configured to determine candidate recommended content that matches the attribute value of the first recommended content under the target attribute category; acquire conversion data of the candidate recommended content in multiple time periods; the amount of conversion data in each time period is negatively correlated with the distance of the time period from the current time.

[0266] In one embodiment, the data acquisition module 802 is further configured to determine the current target attribute category according to the priority order of each target attribute category when there are multiple target attribute categories; if the candidate recommended content under the current target attribute category does not meet the content extraction conditions, then select the next target attribute category as the current target attribute category; until the candidate recommended content under the current target attribute category meets the content extraction conditions, then select the second recommended content from the candidate recommended content under the current target attribute category.

[0267] In one embodiment, the calibration information determination module 806 is used to obtain the amplification factor corresponding to the initial estimated conversion data when the first recommended content is in the initial exposure stage; and to determine the calibration information corresponding to the initial estimated conversion data based on the difference between historical conversion data and estimated conversion data and the amplification factor.

[0268] In one embodiment, the calibration information determination module 806 is used to determine the target resource consumption range corresponding to the first recommended content from the preset resource consumption range; the preset resource consumption range is pre-divided according to the difference between the actual resource consumption value and the target resource consumption value of the first recommended content; and to determine the expansion coefficient corresponding to the target resource consumption range.

[0269] In one embodiment, the calibration information determination module 806 is used to obtain the deviation between the current actual total resource consumption and the current target total resource consumption; the current actual total resource consumption is the total amount of resources currently actually consumed by the first recommended content; the current target total resource consumption is the total amount of target resource consumption corresponding to the first recommended content; and based on the deviation, an expansion coefficient corresponding to the target resource consumption range is determined.

[0270] In one embodiment, the calibration information determination module 806 is used to obtain a smoothing coefficient when the first recommended content is in the fully exposed stage; the smoothing coefficient is determined based on the historical conversion data corresponding to the first recommended content; and to determine calibration information corresponding to the initial estimated conversion data based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data; wherein the smoothed historical conversion data is obtained by smoothing the historical conversion data corresponding to the second recommended content according to the smoothing coefficient; and the smoothed estimated conversion data is obtained by smoothing the estimated conversion data corresponding to the second recommended content according to the smoothing coefficient.

[0271] In one embodiment, the calibration information determination module 806 is used to determine resource consumption deviation information based on the target resource consumption level corresponding to the first recommended content, and the historical conversion data and estimated conversion data of the second recommended content; adjust the historical conversion data of the second recommended content based on the resource consumption deviation information to obtain adjusted historical conversion data; and smooth the adjusted historical conversion data corresponding to the second recommended content according to a smoothing coefficient to obtain smoothed historical conversion data.

[0272] In one embodiment, the calibration information determination module 806 is used to extract historical conversion data and estimated conversion data under the target resource consumption level from the historical conversion data and estimated conversion data of the second recommended content according to the target resource consumption level corresponding to the first recommended content; and to determine resource consumption deviation information based on the difference between the historical conversion data and estimated conversion data under the target resource consumption level and the historical conversion data and estimated conversion data of the second recommended content.

[0273] In one embodiment, the conversion prediction module 802 is further configured to obtain content attribute information and recommendation object information corresponding to the first recommended content; extract content features from the content attribute information and object features from the recommendation object information through the trained conversion prediction model; and perform conversion prediction on the first recommended content based on the content features and object features to obtain the initial predicted conversion data of the first recommended content.

[0274] In one embodiment, such as Figure 9 As shown, the aforementioned recommended content processing device 800 further includes a model training module 801, used to acquire sample content and training labels; the sample content includes historical recommended content, and the training labels are determined based on historical conversion data and historical access data corresponding to the historical recommended content; through the conversion prediction model to be trained in this round, the sample content features and sample object features corresponding to the sample content are extracted; based on the sample content features and sample object features, the conversion prediction of the sample content is performed to obtain the sample prediction conversion data; based on the difference between the sample prediction conversion data and the training labels, the parameters of the conversion prediction model are adjusted, the next round is used as the current round and training continues until the training conditions are met and training stops.

[0275] Specific limitations regarding the processing device for recommended content can be found in the limitations on the processing method for recommended content described above, and will not be repeated here. Each module in the aforementioned processing device for recommended content can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the operations corresponding to each module.

[0276] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the computer program is executed by the processor, it implements a recommended content processing method.

[0277] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for processing recommended content. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad located on the computer device casing, or an external keyboard, touchpad, or mouse.

[0278] Those skilled in the art will understand that Figure 10 and Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0279] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0280] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0281] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps in the above method embodiments.

[0282] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0283] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0284] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for processing recommended content, characterized in that, The method includes: The conversion rate of the first recommended content is estimated to obtain the initial estimated conversion data of the first recommended content; Obtain historical conversion data and estimated conversion data corresponding to the second recommended content; the second recommended content matches the attribute value of the first recommended content under the target attribute category. When the first recommended content is in the initial exposure stage, the target resource consumption range corresponding to the first recommended content is determined from the preset resource consumption range; the preset resource consumption range is pre-divided according to the difference between the actual resource consumption value and the target resource consumption value of the first recommended content. Obtain the deviation between the current actual total resource consumption and the current target total resource consumption; the current actual total resource consumption is the total amount of resources currently actually consumed by the first recommended content; the current target total resource consumption is the target total resource consumption corresponding to the first recommended content. Based on the deviation, determine the expansion coefficient corresponding to the target resource consumption range; Based on the difference between the historical conversion data and the estimated conversion data, and the amplification factor, determine the calibration information corresponding to the initial estimated conversion data; The initial estimated conversion data is adjusted based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

2. The method according to claim 1, characterized in that, The acquisition of historical conversion data and estimated conversion data corresponding to the second recommended content includes: Obtain conversion data corresponding to candidate recommended content; the candidate recommended content is recommended content that matches the attribute value of the first recommended content under the target attribute category; the conversion data includes historical conversion data and estimated conversion data; From the conversion data, extract the historical conversion data and estimated conversion data corresponding to the second recommended content; the second recommended content is the recommended content that meets the preset screening conditions among the candidate recommended content.

3. The method according to claim 2, characterized in that, The step of obtaining the conversion data corresponding to the candidate recommendation content includes: Identify candidate recommended content that matches the attribute value of the first recommended content under the target attribute category; The conversion data of the candidate recommended content is obtained over multiple time periods; the amount of conversion data in each time period is negatively correlated with the distance of the time period from the current time.

4. The method according to claim 2, characterized in that, Before extracting the historical conversion data and estimated conversion data corresponding to the second recommended content from the conversion data, the method further includes: When there are multiple target attribute categories, the current target attribute category is determined according to the priority order of each target attribute category. If the candidate recommended content under the current target attribute category does not meet the content extraction conditions, then the next target attribute category is selected as the current target attribute category; The process continues until the candidate recommended content under the current target attribute category meets the content extraction conditions, at which point the second recommended content is selected from the candidate recommended content under the current target attribute category.

5. The method according to claim 1, characterized in that, The step of determining calibration information corresponding to the initial estimated conversion data based on the difference between the historical conversion data and the estimated conversion data includes: When the first recommended content is in a fully exposed stage, a smoothing coefficient is obtained; the smoothing coefficient is determined based on the historical conversion data corresponding to the first recommended content. Based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data, the calibration information corresponding to the initial estimated conversion data is determined; The smoothed historical conversion data is obtained by smoothing the historical conversion data corresponding to the second recommended content according to the smoothing coefficient; the smoothed estimated conversion data is obtained by smoothing the estimated conversion data corresponding to the second recommended content according to the smoothing coefficient.

6. The method according to claim 5, characterized in that, The method further includes: Based on the target resource consumption level corresponding to the first recommended content, as well as the historical conversion data and the estimated conversion data of the second recommended content, resource consumption deviation information is determined; Based on the resource consumption deviation information, the historical conversion data of the second recommended content is adjusted to obtain the adjusted historical conversion data; According to the smoothing coefficient, the adjusted historical conversion data corresponding to the second recommended content is smoothed to obtain smoothed historical conversion data.

7. The method according to claim 6, characterized in that, The step of determining resource consumption deviation information based on the target resource consumption level corresponding to the first recommended content, the historical conversion data of the second recommended content, and the estimated conversion data includes: Based on the target resource consumption level corresponding to the first recommended content, extract the historical conversion data and the estimated conversion data under the target resource consumption level from the historical conversion data of the second recommended content and the estimated conversion data. Based on the differences between the historical and estimated conversion data under the target resource consumption level and the historical and estimated conversion data of the second recommended content, resource consumption deviation information is determined.

8. The method according to any one of claims 1 to 7, characterized in that, The step of performing conversion prediction on the first recommended content to obtain initial predicted conversion data for the first recommended content includes: Obtain the content attribute information and recommendation object information corresponding to the first recommended content; The content features in the content attribute information and the object features in the recommendation object information are extracted using the trained conversion prediction model. Based on the content features and the object features, the conversion prediction of the first recommended content is performed to obtain the initial estimated conversion data of the first recommended content.

9. The method according to claim 8, characterized in that, The conversion prediction model is obtained through a model training step, which includes: Obtain sample content and training tags; the sample content includes historical recommended content, and the training tags are determined based on the historical conversion data and historical access data corresponding to the historical recommended content; Using the conversion prediction model to be trained in this round, extract the sample content features and sample object features corresponding to the sample content; Based on the characteristics of the sample content and the characteristics of the sample object, the conversion prediction of the sample content is performed to obtain the sample conversion prediction data; Based on the difference between the sample predicted conversion data and the training labels, the parameters of the conversion prediction model are adjusted, the next round is used as the current round and training continues until the training conditions are met and training stops.

10. A processing apparatus for recommended content, characterized in that, The device includes: The conversion prediction module is used to predict the conversion of the first recommended content and obtain the initial predicted conversion data of the first recommended content. The data acquisition module is used to acquire historical conversion data and estimated conversion data corresponding to the second recommended content; the second recommended content is recommended content that matches the attribute value of the first recommended content under the target attribute category; The calibration information determination module is used to determine the target resource consumption range corresponding to the first recommended content from a preset resource consumption range when the first recommended content is in the initial exposure stage; the preset resource consumption range is pre-divided based on the difference between the actual resource consumption value and the target resource consumption value of the first recommended content; obtain the deviation between the current actual total resource consumption and the current target total resource consumption; the current actual total resource consumption is the total amount of resources currently actually consumed by the first recommended content; the current target total resource consumption is the current target total resource consumption corresponding to the first recommended content; determine the expansion coefficient corresponding to the target resource consumption range based on the deviation; and determine the calibration information corresponding to the initial estimated conversion data based on the difference between the historical conversion data and the estimated conversion data and the expansion coefficient. The data calibration module is used to adjust the initial estimated conversion data based on the calibration information to obtain the conversion prediction result corresponding to the first recommended content.

11. The processing apparatus for recommended content according to claim 10, characterized in that, The data acquisition module is further configured to acquire conversion data corresponding to candidate recommended content; the candidate recommended content is recommended content that matches the attribute value of the first recommended content under the target attribute category; the conversion data includes historical conversion data and estimated conversion data; from the conversion data, historical conversion data and estimated conversion data corresponding to the second recommended content are extracted; the second recommended content is recommended content among the candidate recommended content that meets the preset filtering conditions.

12. The processing apparatus for recommended content according to claim 11, characterized in that, The data acquisition module is further configured to determine candidate recommended content that matches the attribute value of the first recommended content under the target attribute category; acquire conversion data of the candidate recommended content in multiple time periods; the amount of conversion data in each time period is negatively correlated with the distance of the time period from the current time.

13. The processing apparatus for recommended content according to claim 11, characterized in that, The data acquisition module is further configured to determine the current target attribute category according to the priority order of each target attribute category when there are multiple target attribute categories; if the candidate recommended content under the current target attribute category does not meet the content extraction conditions, then the next target attribute category is selected as the current target attribute category. The process continues until the candidate recommended content under the current target attribute category meets the content extraction conditions, at which point the second recommended content is selected from the candidate recommended content under the current target attribute category.

14. The processing apparatus for recommended content according to claim 10, characterized in that, The calibration information determination module is further configured to obtain a smoothing coefficient when the first recommended content is in a fully exposed stage; the smoothing coefficient is determined based on the historical conversion data corresponding to the first recommended content; and to determine calibration information corresponding to the initial estimated conversion data based on the difference between the smoothed historical conversion data and the smoothed estimated conversion data; wherein the smoothed historical conversion data is obtained by smoothing the historical conversion data corresponding to the second recommended content according to the smoothing coefficient; and the smoothed estimated conversion data is obtained by smoothing the estimated conversion data corresponding to the second recommended content according to the smoothing coefficient.

15. The processing apparatus for recommended content according to claim 14, characterized in that, The calibration information determination module is further configured to determine resource consumption deviation information based on the target resource consumption level corresponding to the first recommended content, the historical conversion data of the second recommended content, and the estimated conversion data; adjust the historical conversion data of the second recommended content based on the resource consumption deviation information to obtain adjusted historical conversion data; and smooth the adjusted historical conversion data corresponding to the second recommended content according to the smoothing coefficient to obtain smoothed historical conversion data.

16. The processing apparatus for recommended content according to claim 15, characterized in that, The calibration information determination module is further configured to extract historical conversion data and estimated conversion data under the target resource consumption level from the historical conversion data and estimated conversion data of the second recommended content according to the target resource consumption level corresponding to the first recommended content; and determine resource consumption deviation information based on the difference between the historical conversion data and estimated conversion data under the target resource consumption level and the historical conversion data and estimated conversion data of the second recommended content.

17. The processing apparatus for recommended content according to any one of claims 10 to 16, characterized in that, The conversion prediction module is further configured to obtain content attribute information and recommendation object information corresponding to the first recommended content; extract content features from the content attribute information and object features from the recommendation object information using a trained conversion prediction model; and perform conversion prediction on the first recommended content based on the content features and the object features to obtain initial predicted conversion data for the first recommended content.

18. The processing apparatus for recommended content according to claim 17, characterized in that, The device further includes a model training module for acquiring sample content and training labels; the sample content includes historical recommended content, and the training labels are determined based on historical conversion data and historical access data corresponding to the historical recommended content; through the conversion prediction model to be trained in this round, sample content features and sample object features corresponding to the sample content are extracted; based on the sample content features and the sample object features, the conversion of the sample content is predicted to obtain sample predicted conversion data; based on the difference between the sample predicted conversion data and the training labels, the parameters of the conversion prediction model are adjusted, the next round is used as the current round and training continues until the training conditions are met and training stops.

19. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 9.

20. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 9.

21. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 9.

Images