Method and device for content recommendation, equipment and storage medium

Abstract

The embodiment of the invention provides a content recommendation method and device, equipment and a storage medium. The method comprises the following steps: for a first recommended content item in a recommended content set, determining a first estimated value measure obtained by putting the first recommended content item; determining a first resource efficiency value of the first recommended content item based on a historical resource allocation amount and a historical value metric associated with the first recommended content item; determining a first resource allocation amount to be allocated to the first recommended content item based on a first resource efficiency value of the first recommended content item and a total resource allocation amount for the recommended content set; determining a first target estimated value measure for the first recommended content item based on the first estimated value measure and the first resource allocation amount; and determining a sorting result of the first recommended content item in the recommended content set based on the first target estimated value measurement, the first recommended content item being delivered based on the sorting result.

Description

Technical Field

[0001] Example embodiments of the present disclosure generally relate to the field of computer technology, and more particularly, to methods, devices, apparatuses, and computer-readable storage media for content recommendation. Background Art

[0002] The Internet provides access to a wide variety of resources. For example, various applications, commodities, audio and video content, etc. can be accessed through the Internet. In addition, the accessible content also includes specific recommended content items related to various objects / resources, such as advertisements. The content recommendation system can provide content delivery parties with information about the delivery of recommended content items, such as placing recommended content items in specific delivery opportunities to obtain corresponding content clicks, conversions, etc. The delivery of recommended content items is usually based on competition. In the recommendation system, for new recommended content items, their recommendation strategies need to be specially designed.

[0003] Taking advertising as an example, advertisers often have new delivery needs during the promotion process. The typical operation is to create a new delivery plan to obtain the target delivery effect. From the customer's perspective, new plans often encounter cold start problems, which are manifested in poor initial effects and unstable costs, affecting advertisers' delivery experience and reducing customer budgets and other investment willingness. From the perspective of the content recommendation system, new ads are completely unfamiliar to each module in the delivery link. For example, the estimation module does not learn sufficiently when the sample size is insufficient. The chain reaction of inaccurate estimates can easily lead to insufficient platform bidding, affecting the content delivery effect and revenue. Summary of the invention

[0004] In a first aspect of the present disclosure, a method for content recommendation is provided. The method includes: determining, for a first recommended content item in a recommended content set, a first estimated value metric obtained by placing the first recommended content item; determining a first resource efficiency value of the first recommended content item based on a historical resource allocation amount and a historical value metric associated with the first recommended content item; determining a first resource allocation amount to be allocated to the first recommended content item based on the first resource efficiency value of the first recommended content item and a total resource allocation amount for the recommended content set; determining a first target estimated value metric for the first recommended content item based on the first estimated value metric and the first resource allocation amount; and determining a ranking result of the first recommended content item in the recommended content set based on the first target estimated value metric, and the first recommended content item is placed based on the ranking result.

[0005] In a second aspect of the present disclosure, a device for content recommendation is provided. The device includes: a value metric module, configured to determine, for a first recommended content item in a recommended content set, a first estimated value metric obtained by the first recommended content item being delivered; an efficiency determination module, configured to determine a first resource efficiency value of the first recommended content item based on a historical resource allocation amount and a historical value metric associated with the first recommended content item; a resource allocation amount determination module, configured to determine a first resource allocation amount to be allocated to the first recommended content item based on the first resource efficiency value of the first recommended content item and the total resource allocation amount for the recommended content set; a target value metric module, configured to determine a first target estimated value metric for the first recommended content item based on the first estimated value metric and the first resource allocation amount; and a ranking determination module, configured to determine a ranking result of the first recommended content item in the recommended content set based on the first target estimated value metric, and the first recommended content item is delivered based on the ranking result.

[0006] In a third aspect of the present disclosure, an electronic device is provided. The device includes at least one processing unit; and at least one memory, the at least one memory is coupled to the at least one processing unit and stores instructions for execution by the at least one processing unit. When the instructions are executed by the at least one processing unit, the device executes the method of the first aspect.

[0007] In a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein a computer program is stored on the medium, and when the computer program is executed by a processor, the method of the first aspect is implemented.

[0008] It should be understood that the content described in this section is not intended to limit the key features or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0010] Figure 1 A schematic diagram showing an example environment in which embodiments of the present disclosure can be implemented;

[0011] Figure 2 A flowchart showing a process for content recommendation according to some embodiments of the present disclosure is shown;

[0012] Figure 3A shows an example of a value function according to some embodiments of the present disclosure;

[0013] Figure 3B An example of a cost function and its approximate function according to some embodiments of the present disclosure is shown;

[0014] Figure 4 An example of an architecture for content recommendation according to some embodiments of the present disclosure is shown;

[0015] Figure 5 A block diagram showing an apparatus for content recommendation according to some embodiments of the present disclosure; and

[0016] Figure 6 A block diagram of an electronic device is shown in which one or more embodiments of the present disclosure may be implemented. DETAILED DESCRIPTION

[0017] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0018] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below.

[0019] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0020] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0021] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information, so that the user can independently choose whether to provide personal information to software or hardware such as electronic devices, applications, servers or storage media that execute operations of the technical solution of the present disclosure based on the prompt message.

[0022] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information is sent to the user in a manner such as a pop-up window, in which the prompt information can be presented in text form. In addition, the pop-up window can also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0023] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0024] Figure 1 A schematic diagram of an example environment 100 in which embodiments of the present disclosure can be implemented is shown. A content recommendation system 110 is configured to present recommended content items in a placement opportunity. The recommended content items to be recommended may include, for example, one or more recommended content items 122-1, 122-2, ... 122-M (collectively or individually referred to as recommended content items 122 for ease of discussion) in a content database 120.

[0025] Recommended content items can be presented to individual users. Figure 1 As shown, one or more terminal devices 130-1, 130-2, 130-3, etc. (for ease of discussion, collectively or individually referred to as terminal devices 130) are associated with the content recommendation system 110, and can access various types of content provided on the content recommendation system 110, for example, based on corresponding audiences 132-1, 132-2, 132-3, etc. (for ease of discussion, collectively or individually referred to as audiences 132). As an example, the content recommendation system 110 can be an application, a website, a webpage, and other accessible platforms. The terminal device 130 can be installed with an application for accessing the content recommendation system 110, or can access the content recommendation system 110 in an appropriate manner.

[0026] The content recommendation system 110 may be configured to place one or more specific recommended content items in corresponding placement opportunities or content display opportunities (eg, provide or place them in the terminal device 130 ) based on corresponding strategies.

[0027] In this article, a recommended content item refers to content that is presented to recommend a corresponding resource. Examples of recommended content items may include advertisements. In this article, delivery opportunities or content display opportunities may be divided according to the granularity of users (or audiences), and each audience may correspond to a delivery opportunity. In addition to users, delivery opportunities may also be divided according to granularity such as a delivery time, a specific display position on a page, etc.

[0028] In some embodiments, the content recommendation system 110 may recommend corresponding recommended content items on the content recommendation system 110 based on the request of the content provider. In the scenario of advertising delivery, the content provider 150 is sometimes also referred to as an advertiser. In some embodiments, the content provider may also pay the content distributor based on the presentation of the recommended content items and subsequent conversions. In some embodiments, the content recommendation system 110 may determine the recommended content items 122 presented in the corresponding delivery opportunity based on the bidding results.

[0029] In environment 100, terminal device 110 can be any type of mobile terminal, fixed terminal or portable terminal, including mobile phone, desktop computer, laptop computer, notebook computer, netbook computer, tablet computer, media computer, multimedia tablet, personal communication system (PCS) device, personal navigation device, personal digital assistant (PDA), audio / video player, digital camera / camcorder, positioning device, television receiver, radio broadcast receiver, e-book device, game device or any combination of the foregoing, including accessories and peripherals of these devices or any combination thereof. In some embodiments, terminal device 110 can also support any type of interface for users (such as "wearable" circuits, etc.). Content recommendation system 110 can, for example, be various types of computing systems / servers that can provide computing power, including but not limited to mainframes, edge computing nodes, computing devices in cloud environments, and the like.

[0030] It should be understood that the structure and function of the various elements in the environment 100 are described for exemplary purposes only and do not imply any limitation on the scope of the present disclosure.

[0031] As mentioned above, cold start is a necessary stage in the recommended content item delivery plan. Content cold start means that new content is unfamiliar to the recommendation system and has no past information accumulation. It is necessary to accumulate a certain amount of exposure and interaction to collect basic data. This process of accumulating basic data is a cold start. In this application, content cold start may also include advertising cold start. In the advertising scenario, cold start refers to the learning stage from the beginning of advertising delivery to stable delivery. During the cold start stage, advertising delivery data will continue to accumulate so that the content recommendation system can obtain enough data samples to perform stable content recommendations. Recommended content items or advertisements in the cold start stage are sometimes also referred to as cold recommended content items or cold advertisements.

[0032] How to make the newly created recommended content items quickly accumulate conversion samples in a short period of time to pass the cold start period is the guarantee for the long-term stable operation and budget introduction of the content delivery platform. This problem can be optimized from multiple links in the content delivery link, such as the recall stage, the estimation stage, the fine ranking stage, and so on. In the content recommendation system, the recall stage refers to recalling a certain number of recommended content items from the global recommended content item candidates. The estimation stage refers to estimating the cost, click-through rate, conversion rate, etc. of the recommended content items, which can usually be achieved by training an estimation model. The fine ranking stage refers to sorting the recalled recommended content items and placing the top or most ranked recommended content items in specific delivery opportunities.

[0033] The cold start problem is a common research topic in the field of content recommendation. Currently, there are many studies and optimizations on this problem to provide content providers with a better experience. Some optimization strategies include cold path recall, generalized estimation, etc. Some optimization strategies focus on supporting strategies in the fine ranking stage of recommended content items. In competition-based content delivery, for a specific delivery opportunity, the recommended content items are usually ranked based on the estimated revenue to select the highest-ranked recommended content item as the winner of the delivery opportunity. The estimated revenue refers to the revenue that can be obtained if the recommended content item is presented to a specific delivery opportunity.

[0034] In advertising scenarios, ecpm (Expected Cost Per Mile) is usually used as a value measurement indicator for estimated revenue. For a placement opportunity, the ecpm of a recommended content item can be calculated using the following formula: i =bid i *pctr i *pcvr i , where bid i Refers to the bidding of recommended content items for delivery opportunities, pctr i Refers to the expected click-through rate of the recommended content item if it is presented to the placement opportunity, pcvr i Refers to the expected conversion rate of the recommended content item when presented to the placement opportunity. If there are N ads competing for the placement opportunity in the fine ranking stage, the estimated revenue of the jth ad on the placement opportunity i is ecpm ij , then the method for determining the advertisements that can compete for the opportunity to be placed can be expressed as:

[0035]

[0036] For ads in the cold start phase, the content recommendation system will apply support strategies to increase exposure opportunities for ads. The support strategy for cold start is generally that the system uses “additional items” to To improve the ranking of "cold" ads. This can be used to optimize the start-up speed of cold ads and improve the cold start pass rate. In this context, the above-mentioned opportunity competition can be expressed as:

[0037]

[0038] Here, "Additional Items" It can also be called the resource allocation amount of the corresponding recommended content item. The resource allocation amount is additionally added to the corresponding recommended content item, and thus can be called a kind of "support" for the recommended content item.

[0039] From the perspective of content recommendation system, is the revenue loss that the content recommendation system needs to bear, that is, the advertisements competing for the opportunity to be placed Participate in sorting, but the final revenue obtained by the content recommendation platform from the content provider is still ecpm ij , that is, the cost to be paid by the content provider is ecpm ij Obviously, although cold start support can improve advertisers' cold advertising experience, it also requires the content recommendation platform to bear certain revenue losses. In order to avoid excessive revenue losses caused by support, the total resource allocation budget of the support strategy is usually given based on comprehensive consideration of cold start revenue and revenue losses. Therefore, the cold start support strategy can be simply expressed as:

[0040]

[0041] The solution for content recommendation support under the constraint of total resource allocation B can be summarized as follows.

[0042] One solution is a fixed support solution. For each placement opportunity, the resource allocation for cold ads is determined by a fixed coefficient (expressed as fixed_coef), that is, Until the total loss of the content recommendation system's revenue in a period of time (i.e., the sum of the resource allocation) reaches the predetermined budget B for that period of time. This solution is relatively simple and easy to implement, but it has the defects of difficult determination of the fixed coefficient and uneven expenditure of the total resource allocation of the budget.

[0043] Another solution is the dynamic support solution. For each placement opportunity, the support income of cold ads is determined according to the dynamic coefficient, that is, The pacing_coef here is usually a dynamic coefficient calculated by the control algorithm by real-time observation of the total resource allocation expenditure progress. For example, it can be determined based on the classic PID control algorithm. Compared with the above solutions, this solution can spend the total resource allocation more smoothly. However, because all ads share the same pacing_coef at any time, the value difference of different ads in the cold start phase is not considered, resulting in low overall support resource efficiency for cold start.

[0044] Although the above description uses the advertising scenario as an example, similar problems still exist for other recommendation scenarios and other types of recommendation content items.

[0045] For the transition of recommended content items during the cold start phase, it is necessary to consider how to allocate the total resource allocation for support to each recommended content item in the cold start phase, so as to maximize the overall cold start value of the content recommendation system. In addition, it is also necessary to consider how each recommended content item should spend the allocated budget in order to quickly pass the cold start phase and improve the customer's delivery experience.

[0046] In the embodiments of the present disclosure, a cold start support scheme based on resource efficiency is proposed, which can dynamically sense resource efficiency in real time and adaptively allocate resources, thereby better responding to changes in the bidding environment such as content provider operations and delivery opportunities, and significantly improving the cold start delivery effect. In some embodiments, the scheme can be implemented in the fine sorting stage of the content recommendation process to optimize the sorting results of each recommended content, thereby affecting whether the recommended content can be presented to a specific delivery opportunity.

[0047] Some example embodiments of the present disclosure will be described below with continued reference to the accompanying drawings.

[0048] Figure 2 FIG. 2 is a flowchart showing a process 200 for content recommendation according to some embodiments of the present disclosure. The process 200 may be implemented in Figure 1 The content recommendation system 110 is provided.

[0049] In block 210 , for a first recommended content item in the recommended content set, the content recommendation system 110 determines a first estimated value metric obtained by delivering the first recommended content item.

[0050] In some embodiments, the recommended content set includes a plurality of recommended content items, which refer to recommended content items to be placed in a specific placement opportunity or content display opportunity. For example, the placement opportunities or content display opportunities in the content recommendation system 110 can be divided based on the granularity of users, or based on a specific time period or a specific display position in a page. In some embodiments, the recommended content set includes at least recommended content items that are determined to be in a cold start phase in the content recommendation system 110. In this article, the recommended content set includes at least "cold" recommended content items that are to share the total resource allocation of the content recommendation system 110 within a certain period of time.

[0051] Embodiments of the present disclosure relate to selecting recommended content items from a recommended content set for delivery. For recommended content items in the cold start phase, the probability of the recommended content item being delivered is increased by providing cold start support with certain resources. Here, it is assumed that the "first recommended content item" discussed here is a recommended content item in the cold start phase, and the recommended content set may also include one or more recommended content items in the cold start phase. It can be understood that although a single recommended content item is described below, the content recommendation system 110 can perform cold start resource allocation and delivery determination for each recommended content item in the cold start phase in a similar manner.

[0052] In some embodiments, for the first recommended content item, its estimated value metric can be measured, for example, by the expected cost per thousand impressions ecpm. From the perspective of the content provider (e.g., advertiser), ecpm refers to the expected cost that the content provider (e.g., advertiser) has to pay for the delivery of the recommended content item. From the perspective of the content recommendation system, ecpm refers to the revenue that is expected to be obtained from the content provider after the delivery of the recommended content item.

[0053] At block 220 , the content recommendation system 110 determines a first resource efficiency value for the first recommended content item based on the historical resource allocation amount and the historical value metric associated with the first recommended content item.

[0054] As mentioned above, the support problem of recommended content items in the cold start phase can be divided into two sub-problems. The first sub-problem is the resource allocation problem for support, that is, to consider how to divide the total resource allocation to each recommended content item in the cold start phase so as to maximize the overall cold start value of the content recommendation system. The second sub-problem is the resource allocation amount spending problem for recommended content items, that is, how each recommended content item should spend the allocated resource budget in order to quickly pass the cold start phase to improve the customer's delivery experience. Here, resource allocation refers to any resource that can help the recommended content item improve its ranking position. For the recommended content set as a whole, each recommended content item therein shares a resource allocation budget, that is, the total resource allocation amount.

[0055] The budget allocation problem can be defined as: Assuming that the total resource allocation given by the content recommendation system is B, and it is expected to optimize the cold start effect of the content provider so that the content provider's willingness to release content is more sustained, then it is necessary to consider how to allocate the total resource allocation B to the granularity of the recommended content items to maximize the cold start value. This problem can be expressed as the following value function:

[0056]

[0057] where b krepresents the total resource allocation amount allocated to the kth recommended content item among all K recommended content items in the recommended content set. The above formula (4) represents the value function corresponding to the expected K recommended content items when the sum of the resource allocation amounts of all K recommended content items does not exceed the total resource allocation amount B. Maximize the sum. Figure 3A Shows the value function The example of , which can be considered as a two-parameter function with diminishing marginal returns. The problem in equation (4) above needs to be solved by dynamic programming.

[0058] On the other hand, the resource allocation budget partitioning problem for each recommended content item can be defined as: given the total resource allocation amount b of the kth recommended content item itself k , hoping to optimize the conversion of the recommended content item after it is released, so that the recommended content item can pass the cold start period faster and improve the customer's delivery experience. In advertising bidding, this problem can be converted into the incremental revenue of the kth recommended content item for each delivery opportunity (This portion of the income can be considered as a portion of the total resource allocation b k The resource allocation allocated to the delivery opportunity in the recommendation is used to maximize the number of conversions of the recommended content item. This problem can be expressed as follows:

[0059]

[0060] In the above formula (5), x i The value is 1 or 0, indicating that the kth recommended content item is placed in the placement opportunity or not placed in the placement opportunity, that is, only when it is placed, the kth recommended content item will consume a part of the resource allocation. The calculation can be expressed as It can be approximated to cost-per-conversion (CPA) bidding, which can directly optimize the conversion of recommended content items. In actual recommendation scenarios, the parameter α can be adjusted * To adjust According to the above formula (5), for the kth recommended content item, if the total resource allocation amount b allocated to it does not exceed k In this case, it is expected to select appropriate delivery opportunities for delivery to maximize its conversion number, that is, the expected click-through rate (pctr) and the expected conversion rate (pcvr).

[0061] In allocating resources to recommended content items, resource efficiency indicators can also be considered. Resource efficiency indicators can be defined as Resource allocation The target revenue directly brought by target. The higher the efficiency, the more revenue it can bring to the cold start of the recommended content item under the same resource allocation. Take the target revenue target as the value measurement cost calculation for a single delivery opportunity as an example:

[0062] Table 1

[0063]

[0064] In the above table, assuming that for the three recommended content items, their respective estimated value metrics ecpm are 3, 2, and 1, then without cold start support, the recommended content item with ecpm of 3 will be delivered to the current delivery opportunity. However, if support is given to the recommended content item with ecpm of 1 (i.e., a certain amount of resources are allocated), that is, Then the total value metric of the recommended content item is 4, which is higher than the other two recommended content items, so it can be delivered to the current delivery opportunity. From the perspective of the content recommendation system, the direct benefit of resource allocation for the recommended content item with an ecpm of 1 is 1, and the resulting loss of revenue is Then the resource efficiency k=1 / 3.

[0065] The value function can describe the resource efficiency of the recommended content item and define the functional relationship between the "resource allocation amount" and "direct benefit" invested in the recommended content item. Generally speaking, as the amount of resource allocation is invested, the target growth benefit tends to decrease, and is usually defined as follows: Figure 3A The biconvex function shown in FIG. However, in practical applications, the amount of resource allocation will not be infinite, but often limited. Therefore, in some embodiments, based on the needs of practical applications and for the sake of computational efficiency, the value function is approximated and the value function is regarded as a phenomenon function of the amount of resource allocation and the value metric. That is, when the amount of resource allocation is limited, the amount of resource allocation and the value metric can be approximated as being in a linear relationship.

[0066] Figure 3B An example of a cost function 310 according to some embodiments of the present disclosure is shown. * The value of the increase is c * When the amount of resources allocated to support b * When is infinite, the value metric will be maximized. However, considering the limited amount of resource allocation, the resource allocation and value metric can be approximated as a linear relationship, such as Figure 3B The approximate function of the value function is shown in 312. In this case, The target benefit target can be expressed as a value metric ecpm, and k represents the resource efficiency value.

[0067] Considering the linear relationship between resource allocation and value metric, in practical applications, the historical resource allocation and historical value metric of the first recommended content item can be collected, that is, ( ) constitutes a point on the value function. ), the linear type of the value function can be determined, and then the resource efficiency value k of the first recommended content item can be determined. Here, the historical resource allocation amount can include the resource allocation amount allocated to the first recommended content item when the first recommended content item is successfully delivered to a certain delivery opportunity, and the historical value metric can include the direct benefit obtained from the first recommended content item in the delivery, for example, the benefit can be calculated by ecpm.

[0068] According to the above discussion, a preferred solution is to use the recommended content item as the granularity and allocate a respective “total resource allocation amount” b to each recommended content item based on the resource efficiency value. k , and based on this total resource allocation, regulate the support for the recommended content item. However, if the number of recommended content items that the content recommendation system 110 needs to process is large, then the overall workload and complexity of this solution are large, and the granularity of a single recommended content item may face the dilemma of sparse data and untrustworthy value functions. In some embodiments of the present disclosure, by classifying the recommended content items in the cold start phase, the recommended content items in the same class can share the same value function, that is, the same resource efficiency value. In this way, the resource efficiency values ​​of the same class are combined to approximate the theoretical optimal solution based on efficiency allocation budget and dynamic regulation.

[0069] Specifically, in some embodiments, the content recommendation system 110 divides the recommended content items in the recommended content set into multiple categories. Recommended content items under the same category can be approximately regarded as the same recommended content item for calculating the resource efficiency value, which can greatly reduce the complexity of solving and applying the strategy. There are many optional ways to allocate the recommended content items, including rule-based classification, model-based clustering, etc. For example, rule-based classification can include classifying recommended content items with the same conversion definition into the same category. The same conversion definition can include considering activation, downloading, registration or filling out a form for a recommended content item as a conversion action.

[0070] For each of the multiple categories, the content recommendation system 110 can determine the resource efficiency value for the category based on the historical resource allocation amount of the recommended content items that have been allocated to the category and the historical value metric obtained by presenting the recommended content items of the category. Since the same category may include multiple recommended content items, more historical resource allocation amounts and historical value metrics can be obtained. As mentioned above, the historical resource allocation amount and the historical value metric satisfy a value function with a linear relationship. More historical sampling points will help fit a more accurate resource efficiency value k in the linear function.

[0071] Assuming that the first recommended content item is classified into the first category among the multiple categories, the resource efficiency value for the first category can be determined as the first resource efficiency value of the first recommended content item. For other recommended content items, their resource efficiency values ​​can also be determined similarly.

[0072] Continue to refer back Figure 2 At block 230 , the content recommendation system 110 determines a first resource allocation amount to be allocated to the first recommended content item based on the first resource efficiency value of the first recommended content item and the total resource allocation amount for the recommended content set.

[0073] Assume that the total amount of resources to be allocated is B, the recommended content set is divided into N categories, and the corresponding resource efficiency values ​​are represented as type_1, type_2, ..., type_N. The resource efficiency function of each category is defined as Where target[type_i] represents the target benefit of the recommended content item of the i-th category, such as a value metric, and k[type_i] represents the resource efficiency value of the i-th category. represents the resource allocation amount for the recommended content item of the i-th category.

[0074] According to the above formula (4), the resource allocation problem of N categories of recommended content items can be defined as:

[0075]

[0076] According to the above formula (6), in some embodiments, an optimal solution is to allocate the total resource allocation amount B to a type of recommended content item with the highest resource efficiency value k, that is, type * =Argmaxk[type_i].

[0077] In some embodiments, considering the stability of recommended content items of different categories in the system, another method can be applied, that is, the resources used for support are tilted towards recommended content items with higher resource efficiency values. That is, for a category of recommended content items with higher resource efficiency values, more resource allocations can be allocated from the total resource allocation amount B; on the contrary, for a category of recommended content items with lower resource efficiency, less resource allocations can be allocated from the total resource allocation amount B. In some embodiments, the allocation coefficient allocateCoef[type_i] for each category can be determined based on the resource efficiency values ​​of each of the multiple categories, and the allocation coefficient is determined based on the sum of the resource efficiency value of the corresponding category and the resource efficiency values ​​of all categories. For example, for the first recommended content item currently under consideration, the content recommendation system can determine the first allocation coefficient for the first category by comparing the resource efficiency value of the first category with the resource efficiency values ​​of other categories in the multiple categories. For example, for the first category, the allocation coefficient for the first category can be determined based on the ratio of the resource efficiency value of the first category to the sum of the resource efficiency values ​​of the multiple categories. In some embodiments, to avoid the ratio calculated for a certain category being too low, if the ratio is lower than the lower limit of the distribution coefficient, the distribution coefficient for the first category can be determined as the lower limit of the distribution coefficient. If the ratio is not lower than the lower limit of the distribution coefficient, the distribution coefficient for the first category can be determined as the ratio. In some embodiments, the lower limit of the distribution coefficient can be configured by category.

[0078] For example, the partition coefficient is calculated as follows:

[0079]

[0080] In the above formula (7), MiniCoef[type_i] is the lower limit of the allocation coefficient of category type_i.

[0081] Thus, for each category of recommended content items, the resource allocation amount to be allocated to the recommended content items under the category can be determined from the total resource allocation amount B based at least on the allocation coefficient allocateCoef[type_i] corresponding to the category. For example, the total resource allocation amount that can be allocated to the category type_i can be calculated as:

[0082]

[0083] In some embodiments, for the placement of placement opportunity i, if the total resource allocation of the category has been determined, the allocation coefficient allocateCoef[type_i] can be updated in real time by obtaining the resource efficiency value k[type_i] of each category over a period of time. In some embodiments, the total resource allocation of each category can also be controlled in combination with the dynamic control strategy PID The spending progress is adjusted to achieve the purpose of smooth consumption. Such an allocation scheme can be called a dynamic control scheme combined with resource efficiency. Such an allocation scheme can be expressed as follows:

[0084]

[0085]

[0086] According to the above formula (9), for any category, the sum of the resource allocations allocated to the recommended content items of the category that have been placed in various placement opportunities in the historical time period can be calculated as follows: and the total resource allocation allocated to that category Determine the regulation coefficient pacingCoef[type] of the category in the current time period. Then, according to the above formula (10), for a certain recommended content item and a certain delivery opportunity, the resource allocation amount to be allocated to the recommended content item in the delivery opportunity can be determined based on the regulation coefficient pacingCoef[type] of the category and the expected click rate and expected conversion rate of the recommended content item on the delivery opportunity. For each recommended content item under the same category, the resource allocation amount to be allocated in each delivery opportunity can be calculated according to the above formulas (9) and (10).

[0087] In some embodiments, in order to reduce the complexity of regulation, a regulation coefficient pacingCoef for the current time period may be calculated for the recommended content set or for the total resource allocation amount B. The regulation coefficient pacingCoef may be based on the sum of the total resource allocation amount and the resource allocation amount allocated to the recommended content items that have been placed in the recommended content set in the historical time period. Such dynamic regulation may be expressed as follows:

[0088]

[0089] The control coefficient pacingCoef calculated according to the above formula (11) can be used to control the resource allocation amount of recommended content items of all categories to determine the resource allocation amount to be allocated to each recommended content item under each category in the current delivery opportunity i. Such a control coefficient pacingCoef can be called a dynamic global control coefficient. In some embodiments, for recommended content items under a certain category, the dynamic global control coefficient pacingCoef and the allocation coefficient allocateCoef[type] for the category are used together to determine the resource allocation amount for each recommended content item under the category, which can be expressed as follows:

[0090]

[0091] It can be seen that in the above formula (12), (pacingCoef*allocateCoef[type]) can approach the theoretical optimal solution pacingCoef[type] in the above formula (9).

[0092] For the first recommended content item currently being considered, which is classified into the first category, the first resource allocation amount to be allocated to the first recommended content item in the delivery for delivery opportunity i can be determined similarly according to the above equations (11) and (12).

[0093] Continue to refer Figure 2 In block 240, the content recommendation system 110 determines a first target estimated value metric for the first recommended content item based on the first estimated value metric and the first resource allocation amount. i is the first estimated value metric, then the first target estimated value metric can be expressed as in is the resource allocation amount to be allocated to the first recommended content item in the delivery for delivery opportunity i, which can be considered as the incremental value of the estimated value metric of the first recommended content item.

[0094] In box 250, the content recommendation system 110 determines the ranking result of the first recommended content item in the recommended content set based on the first target estimated value metric. The ranking result will affect whether the first recommended content item will be selected from the recommended content set for delivery. Specifically, the content recommendation system 110 can calculate the target estimated value metric for each recommended content item in the recommended content set. Through the ranking of the target estimated value metric, the recommended content item to be ultimately delivered can be selected from the recommended content set according to a preset rule, for example, the recommended content item with the highest ranking can be selected to be delivered in the current delivery opportunity.

[0095] Figure 4 An example of an architecture 400 for content recommendation according to some embodiments of the present disclosure is shown. Assume that for the delivery of delivery opportunity i, the recommended content set includes K recommended content items, denoted as ad1, ad2, ..., ad K For these recommended content items, the estimated value metrics ecpm1, ecpm2, ..., ecpm can be calculated respectively. K. In addition, the content recommendation system 110 also classifies the K recommended content items 430, for example, into N categories, and calculates the resource efficiency of each category 440, represented as k[type]. Based on the results of the content classification and the resource efficiency values ​​of each category, the allocation coefficient allocateCoef[type] of each category can be determined. In addition, the content recommendation system 110 can also dynamically calculate the control coefficient pacingCoef of the current time period based on the total resource allocation amount B 410 and the total resource allocation amount allocated in the historical time period 420 through the PID control strategy.

[0096] Based on the regulation coefficient pacingCoef and the allocation coefficient allocateCoef[type] of each category, and combined with the expected click-through rate pctr_i and expected conversion rate pcvr_i of each recommended content item on delivery opportunity i, the resource allocation amount for each recommended content item can be determined. In this way, the target estimated value metric of the K recommended content items can be calculated Based on the ranking results of the target estimated value metrics of the K recommended content items, it is assumed that the recommended content item ad2 with the highest ranking is selected to be placed on the placement opportunity i.

[0097] Figure 4 The content recommendation process can be iteratively executed and updated. After completing the delivery of delivery opportunity i, the resource allocation amount allocated to the recommended content item ad2 can be continuously collected. To update the total amount of allocated resource allocation 420, so as to update the regulation coefficient pacingCoef for the next time period. In addition, the resource allocation amount of the recommended content item ad2 The value metric ecpm2 is also collected to update the resource efficiency value of the corresponding category. In some embodiments, the update of the control coefficient and the resource efficiency value can be updated in real time for each placement opportunity, or can be updated periodically. For example, in an embodiment of periodic update, a time period can be set, and the allocated resource allocation amount is collected during the time period for updating the control coefficient, and the allocated resource allocation amount and the value metric brought are collected for updating the resource efficiency value.

[0098] According to the embodiments of the present disclosure, in a cold start support scenario with a given total resource allocation, resource efficiency is introduced to maximize the resource efficiency of recommended content items to achieve value maximization. Moreover, in some embodiments, a practical solution for the optimal solution of value maximization is also provided, which reduces the computational complexity of the content recommendation process and makes it easier to implement.

[0099] Figure 5A block diagram of an apparatus for content recommendation according to some embodiments of the present disclosure is shown. The apparatus 500 may be implemented as or included in the content recommendation system 110. Each module / component in the apparatus 500 may be implemented by hardware, software, firmware, or any combination thereof.

[0100] As shown in the figure, the device 500 includes a value metric module 510, which is configured to determine, for a first recommended content item in a recommended content set, a first estimated value metric obtained by the first recommended content item being delivered. The device 500 includes an efficiency determination module 520, which is configured to determine a first resource efficiency value of the first recommended content item based on a historical resource allocation amount and a historical value metric associated with the first recommended content item. The device 500 also includes a resource allocation amount determination module 530, which is configured to determine a first resource allocation amount to be allocated to the first recommended content item based on the first resource efficiency value of the first recommended content item and the total resource allocation amount for the recommended content set. The device 500 also includes a target value metric module 540, which is configured to determine a first target estimated value metric for the first recommended content item based on the first estimated value metric and the first resource allocation amount; and a ranking determination module 550, which is configured to determine a ranking result of the first recommended content item in the recommended content set based on the first target estimated value metric, and the first recommended content item is delivered based on the ranking result.

[0101] In some embodiments, the allocation determination module 530 includes: an allocation coefficient determination module, configured to determine a first allocation coefficient for a first recommended content item based on a first resource efficiency value of the first recommended content item; and an allocation coefficient-based allocation determination module, configured to determine a first resource allocation amount to be allocated to the first recommended content item based at least on the first allocation coefficient and an expected click-through rate and an expected conversion rate for the first recommended content item.

[0102] In some embodiments, the allocation determination module 530 also includes: a regulation coefficient determination module, configured to determine the regulation coefficient for the current time period based on the sum of the total resource allocation amount and the resource allocation amounts allocated to the recommended content items that have been delivered in the recommended content set within the historical time period; and an allocation determination module based on the regulation coefficient, configured to determine the first resource allocation amount to be allocated to the first recommended content item also based on the regulation coefficient.

[0103] In some embodiments, the efficiency determination module 520 includes: a content classification module, configured to divide the recommended content items in the recommended content set into multiple categories, and the first recommended content item is divided into a first category among the multiple categories; a category efficiency determination module, configured to determine, for each category among the multiple categories, a resource efficiency value for the category based on a historical resource allocation amount allocated to the recommended content items in the category and a historical value measurement obtained from the presentation of the recommended content items in the category; and a content item efficiency determination module, configured to determine the resource efficiency value for the first category as a first resource efficiency value for the first recommended content item.

[0104] In some embodiments, the historical resource allocation amount and the historical value metric satisfy a value function having a linear relationship.

[0105] In some embodiments, the allocation determination module 530 includes: an allocation coefficient determination module, configured to determine a first allocation coefficient for the first category by comparing the resource efficiency value of the first category with the resource efficiency values ​​of other categories in a plurality of categories; and an allocation coefficient-based allocation determination module, configured to determine a first resource allocation amount to be allocated to the first recommended content item from the total resource allocation amount based at least on the first allocation coefficient.

[0106] In some embodiments, the allocation coefficient based allocation determination module is configured to determine the allocation coefficient for the first category based on a ratio of the resource efficiency value of the first category to a sum of resource efficiency values ​​of multiple categories.

[0107] In some embodiments, the allocation determination module based on the allocation coefficient is further configured to determine the allocation coefficient for the first category as the allocation coefficient lower limit if the ratio is lower than the allocation coefficient lower limit; and to determine the allocation coefficient for the first category as the ratio if the ratio is not lower than the allocation coefficient lower limit.

[0108] Figure 6 1 shows a block diagram of an electronic device 600 in which one or more embodiments of the present disclosure may be implemented. It should be understood that Figure 6 The electronic device 600 shown is merely exemplary and should not constitute any limitation on the functionality and scope of the embodiments described herein. Figure 6 The electronic device 600 shown can be used to implement Figure 1 The content recommendation system 110 may be used to implement Figure 5 Device 500.

[0109] like Figure 6As shown, the electronic device 600 is in the form of a general-purpose computing device. The components of the electronic device 600 may include, but are not limited to, one or more processors or processing units 610, a memory 620, a storage device 630, one or more communication units 640, one or more input devices 650, and one or more output devices 660. The processing unit 610 may be an actual or virtual processor and is capable of performing various processes according to a program stored in the memory 620. In a multi-processor system, multiple processing units execute computer executable instructions in parallel to improve the parallel processing capabilities of the electronic device 600.

[0110] The electronic device 600 typically includes a plurality of computer storage media. Such media may be any available media accessible to the electronic device 600, including but not limited to volatile and non-volatile media, removable and non-removable media. The memory 620 may be a volatile memory (e.g., a register, a cache, a random access memory (RAM)), a non-volatile memory (e.g., a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. The storage device 630 may be a removable or non-removable medium, and may include a machine-readable medium, such as a flash drive, a disk, or any other medium, which may be capable of being used to store information and / or data (e.g., training data for training) and may be accessed within the electronic device 600.

[0111] The electronic device 600 may further include additional removable / non-removable, volatile / non-volatile storage media. Figure 6 As shown in , a disk drive for reading or writing from a removable, non-volatile disk (e.g., a "floppy disk") and an optical drive for reading or writing from a removable, non-volatile optical disk may be provided. In these cases, each drive may be connected to the bus (not shown) by one or more data media interfaces. The memory 620 may include a computer program product 625 having one or more program modules that are configured to perform various methods or actions of various embodiments of the present disclosure.

[0112] The communication unit 640 implements communication with other electronic devices through a communication medium. Additionally, the functions of the components of the electronic device 600 can be implemented with a single computing cluster or multiple computing machines that can communicate through a communication connection. Therefore, the electronic device 600 can operate in a networked environment using a logical connection with one or more other servers, a network personal computer (PC), or another network node.

[0113] The input device 650 may be one or more input devices, such as a mouse, a keyboard, a tracking ball, etc. The output device 660 may be one or more output devices, such as a display, a speaker, a printer, etc. The electronic device 600 may also communicate with one or more external devices (not shown) through the communication unit 640 as needed, such as a storage device, a display device, etc., communicate with one or more devices that allow a user to interact with the electronic device 600, or communicate with any device that allows the electronic device 600 to communicate with one or more other electronic devices (e.g., a network card, a modem, etc.). Such communication may be performed via an input / output (I / O) interface (not shown).

[0114] According to an exemplary implementation of the present disclosure, a computer-readable storage medium is provided, on which computer-executable instructions are stored, wherein the computer-executable instructions are executed by a processor to implement the method described above. According to an exemplary implementation of the present disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the method described above.

[0115] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices, equipment, and computer program products implemented according to the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0116] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0117] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, so that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0118] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple implementations of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of a module, program segment or instruction includes one or more executable instructions for realizing the logical function of the specification. In some implementations as replacements, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be realized by a special hardware-based system that performs the function or action of the specification, or can be realized by a combination of special hardware and computer instructions.

[0119] The above descriptions of various implementations of the present disclosure are exemplary, non-exhaustive, and not limited to the disclosed implementations. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The selection of terms used herein is intended to best explain the principles of the implementations, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the various implementations disclosed herein.

Claims

1. A content recommendation method, comprising: For a first recommended content item in the recommended content set, determining a first estimated value metric obtained by delivering the first recommended content item; determining a first resource efficiency value for the first recommended content item based on a historical resource allocation amount and a historical value metric associated with the first recommended content item; determining a first resource allocation amount to be allocated to the first recommended content item based on a first resource efficiency value of the first recommended content item and a total resource allocation amount for the recommended content set; determining a first target estimated value metric for the first recommended content item based on the first estimated value metric and the first resource allocation amount; as well as Based on the first target estimated value metric, a ranking result of the first recommended content item in the recommended content set is determined, and the first recommended content item is delivered based on the ranking result.

2. The method of claim 1 , wherein determining a first resource allocation amount to be allocated to the first recommended content item comprises: determining a first allocation coefficient for the first recommended content item based on a first resource efficiency value of the first recommended content item; as well as A first resource allocation amount to be allocated to the first recommended content item is determined based at least on the first allocation coefficient and an expected click rate and an expected conversion rate for the first recommended content item.

3. The method of claim 2, wherein determining a first resource allocation amount to be allocated to the first recommended content item further comprises: Determining a control coefficient for a current time period based on the total resource allocation amount and the sum of the resource allocation amounts allocated to the recommended content items in the recommended content set that have been delivered in a historical time period; as well as A first resource allocation amount to be allocated to the first recommended content item is also determined based on the regulation coefficient.

4. The method according to claim 1, wherein determining a first resource efficiency value of the first recommended content item comprises: dividing the recommended content items in the recommended content set into a plurality of categories, wherein the first recommended content item is divided into a first category among the plurality of categories; For each of the plurality of categories, determining a resource efficiency value for the category based on historical resource allocations that have been allocated to recommended content items of the category and historical value metrics obtained from presentations of recommended content items of the category; as well as The resource efficiency value for the first category is determined as a first resource efficiency value for the first recommended content item. The method according to claim 4 , wherein the historical resource allocation amount and the historical value metric satisfy a value function having a linear relationship.

6. The method of claim 4, wherein determining a first resource allocation amount to be allocated to the first recommended content item comprises: determining a first allocation coefficient for the first category by comparing the resource efficiency value of the first category with resource efficiency values ​​of other categories in the plurality of categories; as well as A first resource allocation amount to be allocated to the first recommended content item in the delivery is determined from the total resource allocation amount based at least on the first allocation coefficient.

7. The method of claim 6, wherein determining a first allocation coefficient for the first category comprises: An allocation coefficient for the first category is determined based on a ratio of the resource efficiency value of the first category to a sum of the resource efficiency values ​​of the plurality of categories.

8. The method of claim 7, wherein determining a first allocation coefficient for the first category based on the ratio comprises: If the ratio is lower than the lower limit of the partition coefficient, determining the partition coefficient for the first category as the lower limit of the partition coefficient; as well as If the ratio is not lower than the distribution coefficient lower limit, the distribution coefficient for the first category is determined as the ratio.

9. A device for content recommendation, comprising: A value metric module, configured to determine, for a first recommended content item in a recommended content set, a first estimated value metric obtained by placing the first recommended content item; an efficiency determination module configured to determine a first resource efficiency value of the first recommended content item based on a historical resource allocation amount and a historical value metric associated with the first recommended content item; a resource allocation amount determining module configured to determine a first resource allocation amount to be allocated to the first recommended content item based on a first resource efficiency value of the first recommended content item and a total resource allocation amount for the recommended content set; a target value metric module configured to determine a first target estimated value metric for the first recommended content item based on the first estimated value metric and the first resource allocation amount; as well as The ranking determination module is configured to determine a ranking result of the first recommended content item in the recommended content set based on the first target estimated value metric, and the first recommended content item is delivered based on the ranking result.

10. An electronic device comprising: at least one processing unit; as well as At least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions causing the device to perform the method according to any one of claims 1 to 8 when executed by the at least one processing unit.

11. A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method according to any one of claims 1 to 8 when executed by a processor.

Images