Recommended methods, systems, electronic devices, and computer-readable storage media

Abstract

The application discloses a recommendation method, a system, a server, an electronic device and a computer readable storage medium. The method comprises: constructing a recommendation scheme list, the recommendation scheme list comprising N candidate recommendation schemes, and a preset number ratio being satisfied between preset recommendation schemes of M categories in the N candidate recommendation schemes; in response to receiving a recommendation request of a user, obtaining a first recommendation scheme for the user; querying a first candidate recommendation scheme with the same category as the first recommendation scheme in the recommendation scheme list; and in response to querying the first candidate recommendation scheme in the recommendation scheme list, recommending the first candidate recommendation scheme to the user and deleting the first candidate recommendation scheme from the recommendation scheme list. The method solves the technical problem that the existing recommendation method causes a large proportion deviation between recommendation schemes of various categories, a long balance time and an unpredictable problem due to post-position adjustment.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a recommendation method, a recommendation server, a recommendation system, an electronic device, and a computer-readable storage medium. Background Technology

[0002] With the development of information technology and the internet, information overload has become a challenge faced by both consumers and producers. Based on this, personalized recommendation has emerged. The core of personalized recommendation is to recommend content (such as products, information, mechanisms, etc.) that each user may be interested in. Currently, a recommendation campaign often involves multiple recommendation schemes. In the process of personalized recommendation, controlling the recommendation ratio of each scheme is a necessary condition for maintaining a long-term economic ecosystem.

[0003] Existing recommendation methods primarily employ reactive adjustments to the proportions of recommended schemes. This means that after a period of operation, the floating coefficient is adjusted based on the actual proportions of each scheme to bring them closer to a predetermined ratio. While these reactive adjustments can eventually achieve equilibrium, the required time to reach balance is unpredictable, and significant deviations are difficult to avoid at any point within that equilibrium period.

[0004] Therefore, existing recommendation methods suffer from technical problems such as large proportional deviations between recommended schemes due to post-adjustment, long balancing times, and unpredictability. Summary of the Invention

[0005] This application provides a recommendation method, a recommendation server, a recommendation system, an electronic device, and a computer-readable storage medium to solve the technical problems of existing recommendation methods, such as large ratio deviations between recommendation schemes, long balancing times, and unpredictable processes caused by post-adjustment.

[0006] This application provides a recommended method, the method comprising:

[0007] Construct a list of recommended solutions, which includes N candidate recommended solutions. Among the N candidate recommended solutions, M preset recommended solutions satisfy a preset quantity ratio, where N and M are both positive integers greater than 1.

[0008] In response to a recommendation request received from a user, obtain a first recommendation scheme for the user;

[0009] Search the list of recommended solutions for a first candidate recommended solution that is of the same category as the first recommended solution.

[0010] In response to finding the first candidate recommendation in the recommendation list, the first candidate recommendation is recommended to the user, and the first candidate recommendation is deleted from the recommendation list.

[0011] This application embodiment also provides a recommendation server, which includes: a recommendation scheme list construction unit, a recommendation scheme acquisition unit, a candidate recommendation scheme query unit, and a recommendation unit;

[0012] The recommendation scheme list construction unit is used to construct a recommendation scheme list, which includes N candidate recommendation schemes. Among the N candidate recommendation schemes, the preset number ratios of the preset recommendation schemes of M categories meet the preset quantity ratio, where N and M are both positive integers greater than 1.

[0013] The recommendation scheme acquisition unit is used to acquire a first recommendation scheme for the user in response to receiving a user's recommendation request;

[0014] The candidate recommendation scheme query unit is used to query the recommendation scheme list for a first candidate recommendation scheme that is of the same category as the first recommendation scheme.

[0015] The recommendation unit is configured to, in response to finding the first candidate recommendation scheme in the recommendation scheme list, recommend the first candidate recommendation scheme to the user and delete the first candidate recommendation scheme from the recommendation scheme list.

[0016] This application also provides a recommendation system, which includes: a user terminal, a business server, and the aforementioned recommendation server; wherein,

[0017] The user terminal includes: a first receiving module, a first processing module, and a first sending module;

[0018] The first receiving module is used to receive user operation instructions, wherein the operation instructions are actions that can trigger recommended behaviors;

[0019] The first processing module is used to generate instruction information corresponding to the operation instruction based on the operation instruction;

[0020] The first sending module is used to send the instruction information to the service server;

[0021] The service server includes: a second receiving module, a second processing module, and a second sending module;

[0022] The second receiving module is used to receive the instruction information sent by the user terminal;

[0023] The second processing module is used to generate a recommendation request for the user based on the instruction information;

[0024] The second sending module is used to send the recommendation request to the recommendation server;

[0025] The recommendation server includes: a third receiving module;

[0026] The third receiving module is used to receive the recommendation request sent by the business server.

[0027] This application also provides an electronic device, including: a memory and a processor;

[0028] The memory is used to store one or more computer instructions;

[0029] The processor is configured to execute one or more computer instructions to implement the above method.

[0030] This application also provides a computer-readable storage medium storing one or more computer instructions that, when executed by a processor, perform the above-described method.

[0031] Compared with the prior art, the recommendation method provided in this application arranges the preset recommendation schemes of M categories into a recommendation scheme list including N candidate recommendation schemes according to a preset quantity ratio, searches the recommendation scheme list for a first candidate recommendation scheme that is the same category as the first recommendation scheme for the user to be recommended, recommends the first candidate recommendation scheme to the user to be recommended, and deletes the first candidate recommendation scheme from the recommendation scheme list.

[0032] This recommended approach has the following advantages:

[0033] First, the method recommends a first candidate recommendation scheme to the user to be recommended, which is a candidate recommendation scheme of the same category as the first recommendation scheme. Since the first recommendation scheme is a recommendation scheme specifically for the user to be recommended, the method achieves personalized recommendations for users.

[0034] Secondly, the method recommends the first candidate recommendation scheme to the user to be recommended to the candidate recommendation scheme list, and the first candidate recommendation scheme will be deleted from the recommendation scheme list after it is recommended. The recommendation scheme list is composed of M categories of preset recommendation schemes arranged according to preset quantity ratios. Therefore, the method achieves strict control over the recommendation ratio of each preset recommendation scheme.

[0035] Third, the method recommends the first candidate recommendation scheme to the user to be recommended as a candidate recommendation scheme in the recommendation scheme list, and the first candidate recommendation scheme will be deleted from the recommendation scheme list after it is recommended. Therefore, the number of preset recommendation schemes of a certain category recommended by this method will not exceed the total number of candidate recommendation schemes of that category in the recommendation scheme list, thus reducing the proportional deviation between preset recommendation schemes of different categories.

[0036] Fourth, the method recommends the first candidate recommendation scheme to the user to be recommended as a candidate recommendation scheme in the recommendation scheme list, and the first candidate recommendation scheme will be deleted from the recommendation scheme list after it is recommended. Therefore, by controlling the number N of candidate recommendation schemes included in the recommendation scheme list, the time when the recommendation ratio of each category of preset recommendation schemes reaches the balance can be controlled, and the balance time can be predicted.

[0037] Therefore, the recommendation method provided in this application solves the technical problems of large proportional deviations, long balancing times, and unpredictable results among the recommended schemes caused by the post-adjustment in existing recommendation methods. Attached Figure Description

[0038] Figure 1 This is an application system diagram of a recommended method provided in an embodiment of this application;

[0039] Figure 2 This is a flowchart of the recommended method provided in the first embodiment of this application;

[0040] Figure 3 This is a schematic diagram of the recommended scheme list provided in the first embodiment of this application;

[0041] Figure 4 This is a schematic diagram of another list of recommended solutions provided in the first embodiment of this application;

[0042] Figure 5 This is a schematic diagram of another list of recommended solutions provided in the first embodiment of this application;

[0043] Figure 6 This is a schematic diagram showing the comparison results of the recommended method provided in the first embodiment of this application;

[0044] Figure 7 This is a schematic diagram of the structure of the recommendation server provided in the second embodiment of this application;

[0045] Figure 8 This is a schematic diagram of the structure of the recommendation system provided in the third embodiment of this application;

[0046] Figure 9 This is a schematic diagram of the structure of the electronic device provided in the fourth embodiment of this application. Detailed Implementation

[0047] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0048] The following describes the terms used in the embodiments of this application:

[0049] Recommendation refers to the process of responding to a user's recommendation request and recommending content (such as products, information, mechanisms, etc.) that the user may be interested in based on the user's user characteristics (such as personal information, interests, purchasing behavior, and reading history). Both recommendation and search are methods to help users quickly discover useful information, but recommendation is the process of providing content to users by analyzing their user characteristics when they haven't provided a specific request; while search is the process of providing search results to users when they actively provide accurate keywords.

[0050] A recommendation system is a personalized information system that recommends products, information, and mechanisms that users may be interested in. It acts as a bridge connecting users and resources, proactively recommending resources that a user might be interested in or find useful based on their user characteristics. For example, in the gaming industry, a recommendation system can recommend suitable game items based on a player's characteristics (such as skill level and item usage). Similarly, in e-commerce, a recommendation system can recommend products that users are more likely to purchase based on their user characteristics (such as age, location, income level, and gender).

[0051] Recommendation algorithms are a type of algorithm in computer science that uses mathematical algorithms to predict content that a user might be interested in based on the user's characteristic information. Among them, recommendation models are neural network models that implement recommendation algorithms.

[0052] A recommendation model is a neural network model built based on user characteristics and user needs. It can output personalized recommendations for a user based on the input user characteristics.

[0053] With the development of information and internet technologies, information overload has become a challenge faced by both consumers and producers. Personalized recommendations have emerged as a result, and the economic benefits they bring are increasingly significant. The core of personalized recommendations is to suggest content that each user might be interested in, thereby increasing their purchasing power and improving their user experience. For example, recommending products that online shoppers might be interested in can boost their spending. Similarly, recommending game friends that gamers might be interested in can improve player retention and overall gaming experience.

[0054] Personalized recommendation technology is relatively mature, but a single recommendation campaign rarely employs only one approach; multiple strategies are often used. Controlling the proportion of each strategy is essential for maintaining a sustainable long-term economic ecosystem. For example, in campaign A promoting game items, different discount rates are preset. In the short term, items with lower discount rates will sell better and generate greater economic benefits. However, considering the long-term game economy, the overall discount rate recommendation ratio needs to be strictly controlled. For instance, the recommendation ratio for 90% off items could be set at 25%, 80% off items at 50%, and 70% off items at 25%. Similarly, in campaign B promoting game items, items with different rarities and uses are preset. Recommending rare and useful items to players will obviously lead to better sales than recommending non-rare and / or less versatile items. However, considering the long-term game economy, the proportion of rare items must be controlled. Therefore, an excellent item recommendation system should not only rely on recommending the most popular items that players are most likely to buy to increase player purchasing power and economic benefits. Instead, it should improve the overall purchase rate of items by combining popular items with long-tail items through reasonable gift pack combinations and user guidance, thus maintaining a better game economic ecosystem in the long run.

[0055] Current recommendation methods primarily rely on post-event adjustments to regulate the proportions of different recommendations. Each recommendation is assigned a floating coefficient, and after a period of operation, this coefficient is adjusted based on the actual recommendation proportions of each recommendation to bring them back to a pre-set ratio. For example, if data monitoring reveals that the actual recommendation proportion of recommendation A is too high after a period, the floating coefficient for recommendation A is lowered. Conversely, if monitoring reveals that the actual recommendation proportion of recommendation A is too low after a period, the floating coefficient for recommendation A is raised. This spring-like oscillation, through several post-event adjustments, achieves a balance (i.e., the actual recommendation proportions of each recommendation conform to the pre-set ratio), but requires a long and unpredictable balancing time. Furthermore, even after the balancing time, significant deviations are still possible at any point within that timeframe. For recommendation activities with strict requirements on the proportions of each recommendation and a defined balancing time, existing methods are inadequate.

[0056] Therefore, how to provide a recommendation method that can reduce the proportional deviation between various recommended options and predict the balancing time has become a technical problem that urgently needs to be solved by those skilled in the art.

[0057] In view of this, this application provides a recommendation method that arranges M preset recommendation schemes into a recommendation scheme list containing N candidate recommendation schemes according to a preset quantity ratio. The method then searches the recommendation scheme list for a first candidate recommendation scheme that is in the same category as the first recommendation scheme for the user to be recommended, recommends the first candidate recommendation scheme to the user, and removes the first candidate recommendation scheme from the recommendation scheme list. This method not only achieves personalized recommendations for users but also achieves strict control over the recommendation ratio between preset recommendation schemes of different categories, reduces the ratio deviation between preset recommendation schemes of different categories, and ensures the predictability of the balancing time. The recommendation method provided in this application is applicable to any field requiring a recommendation system, such as the gaming industry and e-commerce.

[0058] The following detailed description, in conjunction with specific embodiments and accompanying drawings, further illustrates the recommendation method, recommendation server, recommendation system, electronic device, and computer-readable storage medium described in this application.

[0059] Figure 1 This is an application system diagram of a recommended method provided in an embodiment of this application. For example... Figure 1As shown, the system includes a user terminal 101 and a server terminal 102. The user terminal 101 and the server terminal 102 communicate with each other via a network. The user terminal 101 can be a touch terminal, such as a smartphone, tablet computer, or personal digital assistant (PDA); it can also be a computer terminal, such as a laptop or desktop computer. There can be one or multiple such devices. The server terminal 102 is used to deploy the recommendation method provided in this application. A user initiates a recommendation request through the user terminal 101 (e.g., logging into a shopping website or a game account). The recommendation request is transmitted to the server terminal 102 via the network. The server terminal 102 generates a recommendation scheme for the user using the recommendation method provided in this application and transmits the recommendation scheme to the user terminal 101 via the network. The server terminal 102 can be a recommendation device for the user terminal 101, or it can be a server that deploys the recommendation method provided in this application. The server can be a standalone server or a server cluster consisting of multiple sub-servers, where each sub-server deploys a module of the recommendation method provided in this application. For example, a recommendation scheme list construction module, a recommendation scheme generation module, etc., or a cloud server, where the recommendation method provided in this application is deployed on a cloud server to provide recommendation services to multiple user terminals 101 at the same time.

[0060] The first embodiment of this application provides a recommendation method that can be applied to the recommendation server of a recommendation system.

[0061] Figure 2 This is a flowchart of the recommended method provided in this embodiment. The following is in conjunction with... Figure 2 The recommended method provided in this embodiment will be described in detail. The embodiments described below are used to explain the technical solutions of this application and are not intended to limit actual use.

[0062] like Figure 2 As shown, the recommended method provided in this embodiment includes the following steps S201 to S204.

[0063] Step S201: Construct a list of recommended solutions, which includes N candidate recommended solutions. Among the N candidate recommended solutions, the preset number ratios of the M categories of preset recommended solutions meet the preset ratio, where N and M are both positive integers greater than 1.

[0064] The preset recommendation scheme is the scheme set up by the operator for the recommended activity. For example, an e-commerce operator might set up three types of coupons for a discount coupon recommended activity: a 10% off coupon, a 20% off coupon, and a 30% off coupon. These three coupons constitute the preset recommendation scheme for that recommended activity. Similarly, a game operator might set up three types of gift packs for a game gift pack recommended activity: a common equipment enhancement material gift pack, a rare equipment enhancement material gift pack, and an epic equipment enhancement material gift pack. These three game gift packs constitute the preset recommendation scheme for that recommended activity.

[0065] The pre-set recommendation scheme is determined by the operator during the recommendation activity design phase. It's based on historical data and guided by the activity's design goals, defining the content to be recommended, and does not rely on precise algorithms. For example, if an e-commerce operator designs a coupon recommendation activity for Singles' Day (November 11th), they can use historical data such as the previous year's Singles' Day user spending, actual coupon usage, and recent user spending patterns as a basis. Guided by the activity's design goals (e.g., promoting sales of affordable goods), they can design various coupon categories to recommend, such as 10% off coupons, 20% off coupons, and 30% off coupons. For example, if a game operator designs a promotional event for equipment enhancement material packs, they can use historical data such as recent (e.g., 1 month, 3 months, or 6 months before the event) player purchases of enhancement materials and the actual enhancement status of player equipment as a basis. Guided by the event's design goals (e.g., increasing the purchase rate of long-tail materials), the operator can design various types of enhancement material packs to be promoted, such as: common enhancement material packs, rare enhancement material packs, and epic enhancement material packs. A common pack might contain one long-tail material, a rare pack two, and an epic pack three. If a player wants to purchase an epic pack to quickly enhance their items, they will also purchase three long-tail materials, increasing the purchase rate of long-tail materials.

[0066] In this embodiment, the number of categories of preset recommendation schemes set by the operator for the recommendation activity is defined as M. A recommendation activity has at least one preset recommendation scheme, but the recommendation method provided in this embodiment is for M preset recommendation schemes with a preset quantity ratio; therefore, M is a positive integer greater than 1. In actual recommendation activities, the number of categories of preset recommendation schemes is also greater than 1. For example, a game item recommendation activity designed for a game may include preset recommendation schemes covering all items set in the game. A single item can be used as a recommendation scheme, or multiple items can be combined into one recommendation scheme. Another example is a video recommendation activity designed for a video push platform, which may include preset recommendation schemes covering all videos to be pushed on the platform. Due to the large number of videos to be pushed, videos of the same type can be used as one recommendation scheme, such as comedy videos as one recommendation scheme, lifestyle videos as one recommendation scheme, and product promotion videos as one recommendation scheme. These recommendation schemes may have multiple videos to be pushed, and there may be overlap between the videos. The specific implementation method is not limited here.

[0067] The candidate recommendation schemes are those that can be recommended to users and are components of the recommendation scheme list. In this embodiment, preset recommendation schemes are arranged in the recommendation scheme list according to a preset quantity ratio to form candidate recommendation schemes. For example, the preset recommendation schemes include scheme X, scheme Y, and scheme Z. Schemes X, Y, and Z are arranged in the recommendation scheme list according to a preset quantity ratio, and each scheme X, each scheme Y, and each scheme Z is a candidate recommendation scheme. In this embodiment, the preset recommendation schemes must be arranged into a recommendation scheme list before the candidate recommendation schemes in the recommendation scheme list can be recommended to the user.

[0068] The preset quantity ratio refers to the proportion of each category of preset recommendation schemes designed by the operator during the recommendation campaign design phase; that is, the proportion of each preset recommendation scheme expected to be actually recommended to users. This preset quantity ratio can be determined by the operator based on their understanding of macro data, combined with the design goals of the recommendation campaign and future expectations, and is not obtained through precise algorithm calculation.

[0069] To maintain a sustainable economic ecosystem, pre-selected recommendations for various categories should be presented to users in predetermined quantities. Let's take a game operator's equipment enhancement material pack recommendation campaign as an example. The campaign offers three categories of pre-selected recommendations: Option X (common equipment enhancement material pack), Option Y (rare equipment enhancement material pack), and Option Z (epic equipment enhancement material pack). The virtual value (equipment upgrade speed) and price of Options X, Y, and Z increase sequentially. In the game, although Option Z is more expensive, its production is controlled (e.g., dungeons producing epic equipment enhancement materials have long cooldowns, and there are purchase limits on epic equipment materials sold in shops, also requiring a cooldown). High-VIP players will buy as many epic equipment enhancement materials as possible. Therefore, while heavily recommending Option Z can quickly increase game revenue in the short term, it can also lead to players upgrading their equipment too rapidly. In the game environment, game operators design expected values ​​for player equipment levels (e.g., how long it takes for players in a certain tier to upgrade common equipment to top-tier equipment). These expected equipment levels are related to game version updates and other plans. In this context, it's necessary to control the recommended proportion of Option Z, that is, to control the number of epic equipment enhancement materials players possess, thereby limiting the speed at which players' equipment upgrades. Therefore, the preset quantity proportions are set by game operators considering the impact of various preset recommended options on the game's economic system and player ability enhancement. For example, the preset quantity proportion for Option X might be set to 25%, Option Y to 50%, and Option Z to 25%.

[0070] The recommended solution list is a queue of N candidate recommended solutions, which is also a queue of solutions expanded from M preset recommended solutions according to a preset ratio. That is, each candidate recommended solution in the recommended solution list belongs to one of the M preset recommended solutions. For example, if the preset recommended solutions include solution X, solution Y, and solution Z, then any candidate recommended solution in the recommended solution list is either solution X, solution Y, or solution Z.

[0071] In this embodiment, the number of candidate recommendation schemes included in the recommendation scheme list is defined as N. These N candidate recommendation schemes are expanded from the preset recommendation schemes of M categories according to a preset quantity ratio. The number of candidate recommendation schemes included in the recommendation scheme list can also be understood as the length of the recommendation scheme list. The length of the recommendation scheme list is usually determined by the operator based on the expected balancing time between the preset recommendation schemes of each category, and the historical or expected number of recommendation requests within the balancing time. For example, for a certain game, all players will send 10,000 recommendation requests per hour, and the game operator expects to ensure that the proportion of preset recommendation schemes of each category does not deviate significantly within a unit of time per hour. Therefore, the length of the constructed recommendation scheme list can be set to approximately 10,000. As another example, for a certain product, users trigger recommendation requests an average of 400 times per minute, and the operator expects the balancing time for preset recommendation schemes of each category to be measured in minutes. Therefore, the length of the constructed recommendation scheme list can be set to approximately 400.

[0072] This embodiment provides an optional method for constructing a list of recommended solutions, including the following steps S201-1 to S201-4.

[0073] Step S201-1: Obtain the preset recommendation schemes for the M categories.

[0074] During the recommendation campaign design phase, the operator uses historical data as a basis and the campaign design goals as a guide to determine the content to be recommended and forms multiple preset recommendation schemes for different categories. The operator can import all the preset recommendation schemes for all categories into the recommendation server, allowing the recommendation server to obtain all the preset recommendation schemes for all categories. In this embodiment, the number of preset recommendation scheme categories is defined as M.

[0075] For example, in a game item recommendation event, the game operator set up three preset recommendation schemes: Scheme X, Scheme Y, and Scheme Z. Scheme X is a combination pack of items 1 and 2; Scheme Y is a combination pack of items 3 and 4; and Scheme Z is a combination pack of items 5 and 6. The preset quantity ratio for Scheme X is set to 25%, for Scheme Y to 50%, and for Scheme Z to 25%. The game operator imports the content and preset quantity ratios of Schemes X, Y, and Z into the recommendation server.

[0076] Step S201-2: Determine the number N of candidate recommendation schemes based on the preset balance time for the preset recommendation schemes for the M categories and the number of times the recommendation request is obtained within the preset balance time.

[0077] The preset balance time is the unit of time required to consume all the candidate recommendation schemes in the recommendation scheme list. After consuming all the candidate recommendation schemes in the recommendation scheme list, the actual recommendation ratio of the M categories of preset recommendation schemes included in the recommendation scheme list will not deviate significantly from the preset quantity ratio; that is, the preset recommendation schemes of the M categories will be balanced. Therefore, in this embodiment, the time required to consume all the candidate recommendation schemes in the recommendation scheme list is defined as the preset balance time. The preset balance time is a parameter pre-set by the operator based on expectations. For example, if the operator expects to ensure that the actual recommendation ratio of each category of preset recommendation schemes does not deviate significantly from the preset quantity ratio within a unit of hour, then the preset balance time can be set to 1 hour. Or, if the operator expects to ensure that the preset recommendation schemes of each category are balanced within a unit of minute, then the preset balance time can be set to 1 minute.

[0078] The number of times the recommendation request is retrieved within the preset balance time can be determined based on the user activity level during historical periods. For example, if an e-commerce operator designs a coupon recommendation campaign for Singles' Day, the operator determines the number of recommendation requests to be retrieved in this year's Singles' Day coupon recommendation campaign based on the user activity level during the previous year's Singles' Day, i.e., the total number of times all users triggered recommendation requests (login to the online shopping page). Assuming that the average number of times all users triggered recommendation requests within one hour during the previous year's Singles' Day was 20,000, and the preset balance time set for this year's Singles' Day coupon recommendation campaign is one hour, then the number of recommendation requests retrieved within the preset balance time can be 20,000. Of course, user data from historical periods this year (e.g., October) can also be used to make a more accurate prediction of the number of recommendation requests retrieved within the preset balance time.

[0079] The operator imports the preset balancing time and the number of times the recommendation request is obtained within the preset balancing time as parameters into the recommendation server. Based on the preset balancing time and the number of times the recommendation request is obtained within the preset balancing time, the recommendation server can determine the number of candidate recommendation schemes. In this embodiment, the number of candidate recommendation schemes in the recommendation scheme list is defined as N.

[0080] In one optional implementation of this embodiment, N is one or more times the number of acquisitions. Continuing with the example above, assuming the game operator sets the preset balancing time to 1 minute, and the number of acquisitions for the recommendation request within the predetermined preset balancing time is 40, then the game server can determine that the number N of candidate recommendation schemes in the recommendation scheme list for this game item recommendation activity is 40, that is, the length of the recommendation scheme list is 40. Of course, the number N of candidate recommendation schemes in the recommendation scheme list can also be determined to be 80, 120, or 160, etc., based on parameters such as the expected expansion information of this recommendation activity input by the game operator.

[0081] Step S201-3: Determine the amplification factor based on the number N of candidate recommended schemes and the preset quantity ratio.

[0082] Once the number of candidate recommended schemes in the recommended scheme list is determined, the amplification factor can be determined according to the preset quantity ratio, that is, how many times the preset recommended schemes of each category will be increased to form the recommended scheme list.

[0083] Continuing with the example above, the number of candidate recommended schemes N in the recommended scheme list is 40. The preset quantity ratio of scheme X is 25%, the preset quantity ratio of scheme Y is 50%, and the preset quantity ratio of scheme Z is 25%. Then, the ratio of the least common divisor of schemes X, Y, and Z is [1:2:1]. That is to say, the minimum recommended scheme group includes 1 scheme X, 2 schemes Y, and 1 scheme Z, for a total of 4 candidate recommended schemes. Expanding 4 candidate recommended schemes to 40 candidate recommended schemes results in an expansion factor of 10.

[0084] Of course, if the number of candidate recommended solutions in the determined recommended solution list is not an integer multiple of the number of candidate recommended solutions included in the minimum recommended solution group, then the number of candidate recommended solutions in the recommended solution list can be appropriately increased or decreased. For example, if the number of candidate recommended solutions N in the determined recommended solution list is 40, but the minimum recommended solution group includes 6 candidate recommended solutions (the ratio of the least common divisor of solutions X, Y, and Z is [2:3:1]), then the number of candidate recommended solutions N in the recommended solution list can be adjusted from 40 to 42 or 36, so that the preset recommended solutions of each category can exist in the recommended solution list in a preset quantity ratio.

[0085] Step S201-4: Amplify the preset recommended schemes for each category according to the amplification factor to form the recommended scheme list.

[0086] Once the expansion factor is determined, the preset recommended schemes for each category can be expanded to form a list of recommended schemes. All recommended schemes included in the list are candidate recommended schemes.

[0087] Continuing with the example above, after determining the amplification factor to be 10, schemes X, Y, and Z are all amplified by 10 times. Scheme X is amplified to 10, scheme Y to 20, and scheme Z to 10. Arranging these 10 schemes X, 20 schemes Y, and 10 schemes Z creates a recommended scheme list, which can be represented as [X, ..., Y, ..., Z, ...]. Each scheme X, Y, and Z in the recommended scheme list is a candidate recommended scheme and can be recommended to the player.

[0088] Figure 3 This is a schematic diagram of the recommended scheme list provided in this embodiment.

[0089] like Figure 3 As shown, in a game item recommendation event, the game operator's preset recommendation schemes include Scheme X, Scheme Y, and Scheme Z. The preset quantity ratio of Scheme X is 25%, Scheme Y is 50%, and Scheme Z is 25%. The preset balancing time is 1 minute, and the preset number of recommendation requests within the balancing time is 40. Therefore, the constructed recommendation scheme list includes 10 Schemes X, 20 Schemes Y, and 10 Schemes Z. The proportions of Schemes X, Y, and Z in the recommendation scheme list conform to the preset quantity ratios. The order of Schemes X, Y, and Z is not limited.

[0090] It should be noted that for a recommendation campaign corresponding to a business application, the recommendation server will construct a list of recommendation schemes once. Whether it is the same recommendation campaign corresponding to different business applications or different recommendation campaigns corresponding to the same business application, the recommendation server will construct a corresponding list of recommendation schemes. Therefore, there is a one-to-one correspondence between the list of recommendation schemes, business applications, and recommendation campaigns.

[0091] Step S202: In response to receiving a user's recommendation request, obtain a first recommendation scheme for the user.

[0092] The recommendation request can be request information generated when a user performs a request operation, such as when a user enters "×× product" on a website's search page through a user terminal and clicks the search button; or it can be request information generated when a user performs other operations, such as when a user logs into a game account through a user terminal.

[0093] Recommendation requests can be sent directly from the user's client to the recommendation server, or they can be forwarded to the recommendation server through an intermediate node. For example, when a user logs into a game account through the client, the client's processing module detects this action, generates a recommendation request, and sends it to the recommendation server. Another example: when a user logs into a game account through the client, the client's processing module detects this action, generates the user's action information, and sends it to the game server. The game server responds to this action information, generates a recommendation request, and sends it to the recommendation server. The specific implementation depends on the specific business application and is not limited here.

[0094] In one optional implementation provided in this embodiment, the recommendation request includes the user's user feature information, and the method for obtaining the recommendation scheme includes the following steps S202-1 to S202-3.

[0095] Step S202-1: Input the user's recommendation request into a preset recommendation model so that the recommendation model can determine multiple recommendation schemes for the user based on the user feature information in the recommendation request.

[0096] The user characteristic information includes user status information and user behavior information. User status information consists of personal information unrelated to specific business applications, such as age, region, income level, and gender. User status information has some guiding significance in recommendation algorithms; for example, the probability of recommending women's clothing to male users is lower, and the probability of recommending extra-thick cotton-padded clothes to users in southern regions is lower. User behavior information refers to user actions related to specific business applications, such as user purchasing behavior and historical browsing history. This information is related to specific business applications; for example, for games, user behavior information might include user skill usage, user scores, and user item purchases; for video applications, user behavior information might include user video viewing and video liking. User behavior information also has some guiding significance in recommendation algorithms; for example, the probability of recommending daggers to users who frequently use laser guns is lower, and the probability of recommending lifestyle videos to users who frequently like funny videos is lower.

[0097] The recommendation model can be a module deployed on a recommendation server or a standalone server. After the recommendation server inputs a recommendation request, including user feature information, into the recommendation model, the model can output personalized recommendation schemes for that user based on their user feature information. Typically, there are multiple recommendation schemes, and the recommendation model assigns a predicted user acceptance probability to each scheme. For a product, the user acceptance probability is the user's purchase probability. For example, the user acceptance probability for recommendation scheme 1 is 0.85, for recommendation scheme 2 it is 0.12, and for recommendation scheme 3 it is 0.40, etc. The recommendation model often outputs one or more high-probability recommendation schemes sequentially based on the user acceptance probabilities of each scheme.

[0098] The method of using recommendation models to output recommendation solutions is relatively mature. A recommendation model is a neural network model that outputs personalized recommendation solutions for users based on user feature information. The specific architecture and processing methods of recommendation models will not be described in detail here.

[0099] Step S202-2: Sort the multiple recommended schemes from high to low according to their expected virtual value.

[0100] The expected virtual value is a reference value for measuring the potential revenue of each recommendation scheme, and it is more effective in increasing recommendation revenue than simply measuring the recommendation scheme based on the predicted user acceptance probability. For products, one possible method for calculating the expected virtual value is as follows:

[0101] Expected virtual value = Probability of purchase × Virtual price of the product

[0102] The recommendation server can rank multiple recommendation options based on their expected virtual value. For example, a recommendation model for an item might output five recommendations for player A based on player characteristics: Item 1, Item 2, Item 3, Item 4, and Item 5. The predicted purchase probabilities for these five options are 0.50, 0.1, 0.25, 0.05, and 0.1, respectively. The in-game prices of these five items are 10 yuan, 5 yuan, 3 yuan, 4 yuan, and 1 yuan, respectively. The expected virtual value of each recommendation option is shown in Table 1.

[0103] Table 1. Ranking of Recommended Solutions

[0104] 1 Props 1 0.50 10 5 2 Props 3 0.25 3 0.75 3 Props 2 0.1 5 0.5 4 Props 4 0.05 4 0.2 5 Props 5 0.1 1 0.1

[0105] As shown in Table 1, the expected virtual value of item 1 is much greater than that of other items. Although the predicted purchase probability of item 5 is greater than that of item 4, the higher price of item 4 makes the expected virtual value of item 4 exceed that of item 5.

[0106] Step S202-3: Output the first recommendation scheme from the multiple recommendation schemes sorted from high to low based on the expected virtual value.

[0107] After the recommendation server sorts the recommended schemes according to their expected virtual value, it can output any one of the recommended schemes according to the sorting order. In this embodiment, the output recommended scheme is defined as the first recommended scheme.

[0108] In one optional implementation of this embodiment, the first recommended solution is the one with the highest expected virtual value among the plurality of recommended solutions. Outputting the first recommended solution from the plurality of recommended solutions sorted from highest to lowest expected virtual value includes: selecting the one with the highest expected virtual value from the plurality of recommended solutions as the first recommended solution for output.

[0109] The recommendation model calculates multiple recommendations based on the input user characteristics, but the final output recommendation is usually only one: the recommendation with the highest probability of user acceptance, purchase, or expected virtual value. In this embodiment, the recommendation server responds to the user's recommendation request, and the first recommendation obtained for that user is the one with the highest expected virtual value, such as item 1 in Table 1.

[0110] Step S203: Query the list of recommended solutions for a first candidate recommended solution that is in the same category as the first recommended solution.

[0111] In existing technologies, after obtaining a first recommendation for a user, the recommendation server directly recommends that first recommendation to the user. However, in the recommendation method provided in this embodiment, the recommendation server queries the recommendation list to see if there are any candidate recommendation schemes of the same category as the first recommendation. In this embodiment, any one of the candidate recommendation schemes of the same category as the first recommendation is defined as the first candidate recommendation.

[0112] like Figure 3 The provided list of recommended solutions includes 10 solutions X, 20 solutions Y, and 10 solutions Z. When the recommendation server receives a user's recommendation request and obtains the first recommended solution A for that user, it will then search the list of recommended solutions for whether there are any candidate recommended solutions of the same category as the first recommended solution A. For example, if solution X is in the same category as the first recommended solution A, then any one of the solutions X in the list can be used as the first candidate recommended solution.

[0113] Step S204: In response to finding the first candidate recommendation scheme in the recommendation scheme list, the first candidate recommendation scheme is recommended to the user, and the first candidate recommendation scheme is deleted from the recommendation scheme list.

[0114] If the recommendation server finds a first candidate recommendation that is in the same category as the first recommendation in the recommendation list, it will recommend the first candidate recommendation to the user and delete the first candidate recommendation from the recommendation list. In other words, the recommendation server will take the first candidate recommendation that is in the same category as the first recommendation from the recommendation list and return it to the user. The first candidate recommendation in the recommendation list is then consumed.

[0115] Figure 4 This is a schematic diagram of another list of recommended solutions provided in this embodiment.

[0116] by Figure 3 Taking the provided list of recommended solutions as an example, it includes 10 solutions X, 20 solutions Y, and 10 solutions Z. When the recommendation server receives a user's recommendation request and obtains the first recommended solution A for that user, it checks the list of recommended solutions to see if there are any candidate recommended solutions of the same category as the first recommended solution A. If solution X is in the same category as the first recommended solution A, and solution X is already included in the list, then the recommendation server will select any one of solution X as the first candidate recommended solution, remove it from the list, and return it to the user. At this point, the list of recommended solutions will look like this. Figure 4 As shown, the candidate recommended schemes include 9 schemes X, 20 schemes Y, and 10 schemes Z.

[0117] As the recommendation server continuously receives recommendation requests from the same or different users, it continuously retrieves candidate recommendations from the recommendation list and returns them to the user. Since retrieving candidate recommendations also involves removing them from the list, the number of candidate recommendations in the list gradually decreases as the recommendation activity progresses. Typically, two scenarios occur: first, all candidate recommendations of one or more types are retrieved, for example, all 10 recommendations X have been retrieved and returned to the user, thus exhausting the list of recommendations X; second, all candidate recommendations of all types are retrieved, for example, all 10 recommendations X, 20 recommendations Y, and 10 recommendations Z have been retrieved and returned to the user, thus exhausting the list of recommendations X, Y, and Z. The recommendation methods for these two scenarios are explained below.

[0118] First, if the recommendation server does not find a first candidate recommendation scheme of the same category as the first recommendation scheme in the recommendation scheme list, the recommendation method provided in this embodiment further includes the following steps S11-S13.

[0119] Step S11: In response to the first candidate recommendation scheme not being found in the recommendation scheme list, a second recommendation scheme for the user is obtained, wherein the second recommendation scheme is of a different category than the first recommendation scheme.

[0120] The second recommended option is any one of the multiple recommended options output by the recommendation model, excluding the first recommended option. In one possible implementation, the second recommended option is any recommended option whose expected virtual value is less than that of the first recommended option. In another possible implementation, the second recommended option is the first recommended option whose expected virtual value is ranked after that of the first recommended option; that is, the expected virtual value of the second recommended option is only less than that of the first recommended option. As shown in Table 1, if item 1 is the first recommended option, then item 3 is the second recommended option.

[0121] If the recommendation server does not find a first candidate recommendation scheme of the same category as the first recommendation scheme in the recommendation scheme list, it means that all candidate recommendation schemes of the same category as the first recommendation scheme in the recommendation scheme list have been retrieved. Then, the recommendation server will re-obtain a second recommendation scheme for the user, such as scheme B, where the expected virtual value of scheme B is less than the expected virtual value of scheme A.

[0122] Step S12: Query the list of recommended solutions for a second candidate recommended solution that is in the same category as the second recommended solution.

[0123] After obtaining the second recommendation solution, the recommendation server will repeat the query operation to find a second candidate recommendation solution in the recommendation solution list that is in the same category as the second recommendation solution. In this embodiment, any candidate recommendation solution in the recommendation solution list that is in the same category as the second recommendation solution but is not in the same category as the first candidate recommendation solution is defined as the second candidate recommendation solution.

[0124] Of course, if no second candidate recommendation scheme of the same category as the second recommendation scheme is found in the recommendation scheme list, the recommendation server will repeat step S11 to obtain a third recommendation scheme for the user. This third recommendation scheme is different in category from both the first and second recommendation schemes. In one optional implementation, the third recommendation scheme is any one of the recommendation schemes whose expected virtual value is less than that of the first and second recommendation schemes. In another optional implementation, the second recommendation scheme is the first recommendation scheme whose expected virtual value ranks after the first recommendation scheme, and the third recommendation scheme is the second recommendation scheme whose expected virtual value ranks after the first recommendation scheme; that is, the expected virtual value of the third recommendation scheme is only less than that of the second recommendation scheme. After obtaining the third recommendation scheme, the recommendation server will repeat step S12 to search the recommendation scheme list for candidate recommendation schemes of the same category as the third recommendation scheme. This process is repeated until no candidate recommendation schemes are found in the recommendation scheme list.

[0125] Step S13: In response to finding the second candidate recommendation scheme in the recommendation scheme list, the second candidate recommendation scheme is recommended to the user, and the second candidate recommendation scheme is deleted from the recommendation scheme list.

[0126] If the recommendation server finds a second candidate recommendation that is in the same category as the second recommendation in the recommendation list, it will recommend the second candidate recommendation to the user and remove it from the recommendation list. In other words, the recommendation server will remove the second candidate recommendation that is in the same category as the second recommendation from the recommendation list and return it to the user.

[0127] Figure 5 This is a schematic diagram of another list of recommended solutions provided in this embodiment.

[0128] by Figure 3Taking the provided list of recommended solutions as an example, it includes 10 solutions X, 20 solutions Y, and 10 solutions Z. As the recommendation activity progresses, the recommendation server retrieves 10 solutions X, 11 solutions Y, and 3 solutions Z from the list and returns them to multiple users. That is, solution X is no longer present in the list, which now includes 9 solutions Y and 7 solutions Z. When the recommendation server receives a user's recommendation request and obtains the first recommended solution A for that user, it checks the list to see if there is a first candidate recommended solution of the same category as solution A. If solution X is in the same category as solution A, but solution X is no longer present in the list, the recommendation server does not find solution X in the list that is in the same category as solution A. The recommendation server then obtains a second recommendation scheme B for the user. It then checks the recommendation scheme list to see if a second candidate recommendation scheme Z exists in the same category as the second recommendation scheme B. If scheme Z is in the same category as the second recommendation scheme B, and the recommendation scheme list also includes scheme Z, then the recommendation server retrieves scheme Z from the recommendation scheme list and returns scheme Z to the user. At this point, the recommendation scheme list will be as follows: Figure 5 As shown, the candidate recommended schemes include 9 schemes Y and 6 schemes Z.

[0129] Second, if the recommendation server does not find a first candidate recommendation scheme of the same category as the first recommendation scheme in the recommendation scheme list, and does not find any candidate recommendation scheme in the recommendation scheme list, the recommendation method provided in this embodiment further includes the following steps S21-S23.

[0130] Step S21: In response to the fact that the first candidate recommendation scheme is not found in the recommendation scheme list and no candidate recommendation scheme is found in the recommendation scheme list, a second recommendation scheme list is generated, wherein the preset number ratios of the preset recommendation schemes of the M categories in the second recommendation scheme list are satisfied.

[0131] If the recommendation server does not find any candidate recommendation scheme in the recommendation scheme list, it means that all candidate recommendation schemes in the list have been consumed. In this case, the recommendation server will regenerate a second recommendation scheme list. It should be noted that the preset quantity ratio among the M categories of preset recommendation schemes in the second recommendation scheme list must still be met. The length of the second recommendation scheme list can be the same as or different from the length of the first recommendation scheme list. For example, the first recommendation scheme list is a queue formed by expanding the preset recommendation schemes of the M categories by 40 times according to the preset quantity ratio, while the second recommendation scheme list is a queue formed by expanding the preset recommendation schemes of the M categories by 50 times according to the preset quantity ratio.

[0132] In one optional implementation provided in this embodiment, the length of the second recommendation scheme list is the same as the length of the original recommendation scheme list; that is, the second recommendation scheme list is identical to the original recommendation scheme list. Typically, for the same recommendation activity within the same business application, regardless of how many recommendation scheme lists are regenerated, each recommendation scheme list will be identical to the original recommendation scheme list.

[0133] The original recommendation scheme list is a list of recommendation schemes constructed by the recommendation server during the recommendation activity design phase, based on imported preset recommendation schemes, preset quantity ratios, preset balancing times, and the number of times recommendation requests are retrieved within the preset balancing time.

[0134] For example: if Figure 3 The list of recommended solutions shown is the original list of recommended solutions. Therefore, the second list of recommended solutions also consists of 10 solutions X, 20 solutions Y, and 10 solutions Z.

[0135] Step S22: Query the second recommendation list for a third candidate recommendation that is in the same category as the first recommendation.

[0136] The recommendation server continues to query candidate recommendation schemes of the same category as the first recommendation scheme in the second recommendation scheme list. In this embodiment, any one of the candidate recommendation schemes of the same category as the first recommendation scheme in the second recommendation scheme list is defined as the third candidate recommendation scheme.

[0137] Step S23: In response to finding the third candidate recommendation scheme in the second recommendation scheme list, the third candidate recommendation scheme is recommended to the user, and the third candidate recommendation scheme is deleted from the second recommendation scheme list.

[0138] If the recommendation server finds a third candidate recommendation in the second recommendation list that is in the same category as the first recommendation, it will recommend the third candidate recommendation to the user and remove it from the second recommendation list. In other words, the recommendation server will remove the third candidate recommendation that is in the same category as the first recommendation from the second recommendation list and return it to the user.

[0139] like Figure 3 The list of recommended solutions shown is the original list. After one round of recommendations, the second list of recommended solutions will look like this. Figure 4 As shown.

[0140] As the recommendation campaign progresses, all candidate recommendation schemes in the second recommendation scheme list will be exhausted. The recommendation server will then regenerate the third recommendation scheme list and repeat the above steps until the recommendation campaign ends.

[0141] In summary, compared with existing post-adjustment recommendation methods, the recommendation method provided in this embodiment is a pre-adjustment method. It first constructs a recommendation scheme list based on the preset quantity ratio of each type of preset recommendation scheme, the preset balancing time, and the number of times recommendation requests are obtained within the preset balancing time. Then, it selects candidate recommendation schemes from the recommendation scheme list and recommends them to the user. It has the advantages of strictly controlling the recommendation ratio between each preset recommendation scheme, reducing the ratio deviation between each preset recommendation scheme, and predicting the balancing time between each preset recommendation scheme.

[0142] Figure 6 This is a schematic diagram comparing the existing recommendation method provided in this embodiment with the recommendation method provided in this embodiment.

[0143] like Figure 6As shown, the horizontal axis (X) represents the running time of a recommendation activity, and the vertical axis (Y) represents the variance of the deviation between the actual recommendation ratio and the preset quantity ratio of each preset recommendation scheme at each running time. Existing recommendation methods, being retrospective adjustments, output recommendation schemes according to inherent parameters. They then monitor the actual recommendation ratio of each scheme with hysteresis and make inverse adjustments, using subsequent recommendation schemes to balance the ratio. Therefore, they exhibit a back-and-forth oscillating deviation curve 601. Through this spring-like oscillation, after several retrospective adjustments, although the recommendation ratio of each preset scheme can be balanced (i.e., the actual recommendation ratio of each preset scheme conforms to the preset quantity ratio), it requires a long balancing time, and the length of this balancing time is unpredictable. The recommendation method provided in this embodiment, being a pre-emptive adjustment, first establishes a list of recommendation schemes that strictly conform to the preset quantity ratio, and then selects candidate recommendation schemes from the list. Therefore, it exhibits a sawtooth-shaped deviation curve 602. As can be seen from deviation curves 601 and 602, the recommendation method provided in this embodiment can reduce the proportional deviation between each preset recommendation scheme and can control the balancing time between each preset recommendation scheme.

[0144] In general, the longer the list of recommended options is, the greater the maximum deviation will be, but it will not exceed the total number of the candidate recommended options with the largest proportion in the list. For example, if the list of recommended options includes 10 options X, 20 options Y, and 10 options Z, then the maximum possible deviation is at most 20 (that is, option Y is recommended to 20 users in a row), and then it will be immediately balanced back.

[0145] The shorter the recommended solution list, the smaller the maximum deviation and the shorter the balancing time. For example, for a certain game, all players send 10,000 recommendation requests per hour. If the recommended solution list is set to 10,000, it will reach balance in a maximum of one hour. If the recommended solution list is set to 20,000, it may take two hours to reach balance. In other words, if all candidate recommended solutions in the recommended solution list are recommended, the recommendation ratio among the preset recommended solutions will inevitably reach equilibrium.

[0146] However, the shorter the recommended option list, the more potential revenue might be lost. For example, suppose a game's equipment enhancement material recommendation event includes options X, Y, and Z, where options X, Y, and Z represent common enhancement materials, rare enhancement materials, and epic enhancement materials, respectively, with preset quantity ratios of 25%, 50%, and 25%. Players are divided into two groups: wealthy players and average players. Wealthy players will buy almost anything recommended, so recommending epic enhancement materials (Option Z) obviously yields the highest revenue. Average players can only afford common enhancement materials (Option X) and rare enhancement materials (Option Y). Therefore, the most profitable strategy would be to allocate the 25% recommendation opportunity for epic enhancement materials (Option Z) to wealthy players. However, in actual game operation, the login and purchasing behavior of wealthy and average players is not continuous and even. It's very possible that wealthy players are much more active during certain periods, and much less active during others. If the recommended options list is short, for example, with only four candidate options: common enhancement materials (Option X), rare enhancement materials (Option Y), and epic enhancement materials (Option Z), then even if four wealthy players trigger recommendation requests consecutively, only one wealthy player will be recommended epic enhancement materials (Option Z), and the purchasing power of the remaining three wealthy players will be wasted. However, if the recommended options list is longer, for example, with 4000 candidate options, of which 1000 are epic enhancement materials (Option Z), then unless a wealthy player triggers recommendation requests more than 1000 times out of the 4000 requests, there will be no waste of purchasing power for wealthy players. The longer the queue, the less the lost benefits.

[0147] Therefore, during the design phase of the recommendation campaign, it is necessary to rationalize the length of the recommendation scheme list to balance the deviations, balancing time, and recommendation benefits among the preset recommendation schemes.

[0148] The second embodiment of this application provides a recommendation server. Figure 7 This is a schematic diagram of the structure of the recommendation server provided in this embodiment.

[0149] like Figure 7 As shown, the recommendation server provided in this embodiment includes: a recommendation scheme list construction unit 701, a recommendation scheme acquisition unit 702, a candidate recommendation scheme query unit 703, and a recommendation unit 704;

[0150] The recommendation scheme list construction unit 701 is used to construct a recommendation scheme list, which includes N candidate recommendation schemes. Among the N candidate recommendation schemes, the preset quantity ratios of the M categories of preset recommendation schemes meet the preset quantity ratios, where N and M are both positive integers greater than 1.

[0151] Optionally, the list of recommended solutions includes:

[0152] Obtain the preset recommendation schemes for the M categories;

[0153] Based on the preset balance time for the preset recommendation schemes for the M categories, and the number of times the recommendation request is obtained within the preset balance time, the number N of the candidate recommendation schemes is determined, wherein the preset balance time is the unit time that represents the consumption of the candidate recommendation schemes in the recommendation scheme list;

[0154] The amplification factor is determined based on the number N of the candidate recommendation schemes and the preset quantity ratio;

[0155] The preset recommended schemes for each category are amplified according to the amplification factor to form the recommended scheme list.

[0156] The recommendation scheme acquisition unit 702 is used to acquire a first recommendation scheme for the user in response to receiving a recommendation request from the user.

[0157] Optionally, the recommendation request includes the user's user characteristic information; the step of obtaining a first recommendation scheme for the user in response to receiving the user's recommendation request includes:

[0158] The user's recommendation request is input into a preset recommendation model, so that the recommendation model determines multiple recommendation schemes for the user based on the user feature information in the recommendation request;

[0159] The multiple recommended schemes are sorted from high to low according to their expected virtual value;

[0160] The first recommendation solution is output from a plurality of recommendation solutions sorted from high to low based on the expected virtual value.

[0161] Optionally, the first recommended solution is the one with the highest expected virtual value among the plurality of recommended solutions; the step of outputting the first recommended solution from the plurality of recommended solutions sorted from high to low expected virtual value includes:

[0162] The recommendation scheme with the highest expected virtual value is selected from the multiple recommendation schemes and output as the first recommendation scheme.

[0163] The candidate recommendation query unit 703 is used to query the recommendation list for a first candidate recommendation scheme that is of the same category as the first recommendation scheme.

[0164] The recommendation unit 704 is configured to, in response to finding the first candidate recommendation scheme in the recommendation scheme list, recommend the first candidate recommendation scheme to the user and delete the first candidate recommendation scheme from the recommendation scheme list.

[0165] Optionally, the recommendation server is also used for:

[0166] In response to the first candidate recommendation not being found in the recommendation list, a second recommendation for the user is obtained, wherein the second recommendation is of a different category than the first recommendation.

[0167] Search the list of recommended solutions for a second candidate recommended solution that is of the same category as the second recommended solution.

[0168] In response to finding the second candidate recommendation in the recommendation list, the second candidate recommendation is recommended to the user and then removed from the recommendation list.

[0169] Optionally, the recommendation server is also used for:

[0170] In response to the fact that no first candidate recommendation scheme is found in the recommendation scheme list and no candidate recommendation scheme is found in the recommendation scheme list, a second recommendation scheme list is generated, wherein the preset number ratios of the preset recommendation schemes of the M categories in the second recommendation scheme list are satisfied.

[0171] Search the second recommended solution list for a third candidate recommended solution that is of the same category as the first recommended solution;

[0172] In response to finding the third candidate recommendation scheme in the second recommendation scheme list, the third candidate recommendation scheme is recommended to the user, and the third candidate recommendation scheme is deleted from the second recommendation scheme list.

[0173] The third embodiment of this application provides a recommendation system. Figure 8 This is a schematic diagram of the recommendation system provided in this embodiment.

[0174] like Figure 8 As shown, the recommendation system provided in this embodiment includes: a user terminal 801, a business server terminal 802, and a recommendation server terminal 803 as described in the second embodiment of this application.

[0175] The user terminal 801 includes: a first receiving module 8011, a first processing module 8012, and a first sending module 8013;

[0176] The first receiving module 8011 is used to receive the user's operation instruction, wherein the operation is an action that can trigger recommended behavior;

[0177] The first processing module 8012 is used to generate instruction information corresponding to the operation instruction according to the operation instruction;

[0178] The first sending module 8013 is used to send the instruction information to the service server 802;

[0179] The service server 802 includes: a second receiving module 8021, a second processing module 8022, and a second sending module 8023;

[0180] The second receiving module 8021 is used to receive the instruction information sent by the user terminal 801;

[0181] The second processing module 8022 is used to generate a recommendation request for the user based on the instruction information;

[0182] The second sending module 8023 is used to send the recommendation request to the recommendation server 803;

[0183] The recommendation server 803 includes: a third receiving module 8031;

[0184] The third receiving module 8031 ​​is used to receive the recommendation request sent by the business server 802;

[0185] The recommendation server 803 further includes a recommendation scheme list construction unit 8032, a recommendation scheme acquisition unit 8033, a candidate recommendation scheme query unit 8034, and a recommendation unit 8035 as described in the second embodiment of this application.

[0186] The recommendation scheme list construction unit 8032 is used to construct a recommendation scheme list, which includes N candidate recommendation schemes. Among the N candidate recommendation schemes, the preset number ratios of the preset recommendation schemes of M categories meet the preset quantity ratio, where N and M are both positive integers greater than 1.

[0187] The recommendation scheme acquisition unit 8033 is used to acquire a first recommendation scheme for the user in response to receiving the recommendation request;

[0188] The candidate recommendation scheme query unit 8034 is used to query the recommendation scheme list for a first candidate recommendation scheme that is the same category as the first recommendation scheme.

[0189] The recommendation unit 8035 is configured to, in response to finding the first candidate recommendation scheme in the recommendation scheme list, recommend the first candidate recommendation scheme to the user and delete the first candidate recommendation scheme from the recommendation scheme list.

[0190] The fourth embodiment of this application provides an electronic device. Figure 9 This is a schematic diagram of the structure of the electronic device provided in this embodiment.

[0191] like Figure 9 As shown, the electronic device provided in this embodiment includes a memory 901 and a processor 902.

[0192] The memory 901 is used to store computer instructions for executing the recommended method.

[0193] The processor 902 is used to execute computer instructions stored in the memory 901, and performs the following operations:

[0194] Construct a list of recommendation schemes, which includes N candidate recommendation schemes. Among the N candidate recommendation schemes, the preset number ratios of the M categories of preset recommendation schemes meet the preset ratios, where N and M are both positive integers greater than 1.

[0195] In response to receiving a user's recommendation request, obtain a first recommendation scheme for the user;

[0196] Search the list of recommended solutions for a first candidate recommended solution that is of the same category as the first recommended solution.

[0197] In response to finding the first candidate recommendation in the recommendation list, the first candidate recommendation is recommended to the user, and the first candidate recommendation is deleted from the recommendation list.

[0198] Optionally, the list of recommended solutions includes:

[0199] Obtain the preset recommendation schemes for the M categories;

[0200] Based on the preset balance time for the preset recommendation schemes for the M categories, and the number of times the recommendation request is obtained within the preset balance time, the number N of the candidate recommendation schemes is determined, wherein the preset balance time is the unit time that represents the consumption of the candidate recommendation schemes in the recommendation scheme list;

[0201] The amplification factor is determined based on the number N of the candidate recommendation schemes and the preset quantity ratio;

[0202] The preset recommended schemes for each category are amplified according to the amplification factor to form the recommended scheme list.

[0203] Optionally, the recommendation request includes the user's user characteristic information; the step of obtaining a first recommendation scheme for the user in response to receiving the user's recommendation request includes:

[0204] The user's recommendation request is input into a preset recommendation model, so that the recommendation model determines multiple recommendation schemes for the user based on the user feature information in the recommendation request;

[0205] The multiple recommended schemes are sorted from high to low according to their expected virtual value;

[0206] The first recommendation solution is output from a plurality of recommendation solutions sorted from high to low based on the expected virtual value.

[0207] Optionally, the first recommended solution is the one with the highest expected virtual value among the plurality of recommended solutions; the step of outputting the first recommended solution from the plurality of recommended solutions sorted from high to low expected virtual value includes:

[0208] The recommendation scheme with the highest expected virtual value is selected from the multiple recommendation schemes and output as the first recommendation scheme.

[0209] Optionally, it can also be used to perform the following operations:

[0210] In response to the first candidate recommendation not being found in the recommendation list, a second recommendation for the user is obtained, wherein the second recommendation is of a different category than the first recommendation.

[0211] Search the list of recommended solutions for a second candidate recommended solution that is of the same category as the second recommended solution.

[0212] In response to finding the second candidate recommendation in the recommendation list, the second candidate recommendation is recommended to the user and then removed from the recommendation list.

[0213] Optionally, it can also be used to perform the following operations:

[0214] In response to the fact that no first candidate recommendation scheme is found in the recommendation scheme list and no candidate recommendation scheme is found in the recommendation scheme list, a second recommendation scheme list is generated, wherein the preset number ratios of the preset recommendation schemes of the M categories in the second recommendation scheme list are satisfied.

[0215] Search the second recommended solution list for a third candidate recommended solution that is of the same category as the first recommended solution;

[0216] In response to finding the third candidate recommendation scheme in the second recommendation scheme list, the third candidate recommendation scheme is recommended to the user, and the third candidate recommendation scheme is deleted from the second recommendation scheme list.

[0217] The fifth embodiment of this application provides a computer-readable storage medium, which includes computer instructions that, when executed by a processor, are used to implement the methods described in the embodiments of this application.

[0218] It should be noted that the relational terms such as "first" and "second" used in this document are only used to distinguish one entity or operation from another, and do not require or imply any actual relationship or order between these entities or operations. Furthermore, "including," "having," "containing," and other similar terms are synonymous, and the conclusion of any one or more items following any of the foregoing words is open-ended; none of the foregoing terms indicates that the one or more items have been exhaustively listed, or are limited to only one or more of the listed items.

[0219] When used herein, unless otherwise expressly stated, the term "or" includes all possible combinations except those that are impractical. For example, if expressed as a database may include A or B, then unless otherwise specified or impractical, it may include database A, or B, or A and B. As a second example, if expressed as a database may include A, B, or C, then unless otherwise specified or impractical, the database may include database A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0220] It is worth noting that the above embodiments can be implemented by hardware or software (program code), or a combination of hardware and software. If implemented by software, it can be stored in the above-described computer-readable medium. When executed by a processor, the software can perform the methods disclosed above. The computing units and other functional units described in this disclosure can be implemented by hardware or software, or a combination of hardware and software. Those skilled in the art will also understand that the above-described multiple modules / units can be combined into one module / unit, and each of the above-described modules / units can be further divided into multiple sub-modules / sub-units.

[0221] In the foregoing detailed description, embodiments have been described with reference to numerous specific details, which may vary depending on the implementation. Certain adaptations and modifications can be made to the embodiments. Other implementations will be readily apparent to those skilled in the art from the specific embodiments disclosed herein. This specification and examples are for illustrative purposes only, and the true scope and essence of this application are defined by the claims. The sequence of steps shown in the figures is also for illustrative purposes only and is not intended to limit to any particular step or order. Therefore, those skilled in the art will recognize that these steps can be performed in different orders when implementing the same method.

[0222] Exemplary embodiments are disclosed in the figures and detailed description of this application. However, many variations and modifications can be made to these embodiments. Accordingly, although specific terminology is used, it is only general and descriptive and not for limiting purposes.

Claims

1. A recommendation method, characterized in that, The method is used to recommend products based on user characteristic information, including user personal information, user interests, user purchasing behavior, and user reading history. A recommendation scheme list is constructed, which includes N candidate recommendation schemes. The N candidate recommendation schemes satisfy a preset quantity ratio among the preset recommendation schemes of M categories. The N candidate recommendation schemes are obtained by expanding the preset recommendation schemes of the M categories according to the preset quantity ratio. The number N of candidate recommendation schemes is determined based on the preset balance time for the preset recommendation schemes of the M categories, and the historical or expected number of times recommendation requests are obtained within the preset balance time. The preset balance time is the unit time that represents the consumption of the candidate recommendation schemes in the recommendation scheme list. Where N and M are both positive integers greater than 1. In response to receiving a user's recommendation request, obtain a first recommendation scheme for the user; Search the list of recommended solutions for a first candidate recommended solution that is of the same category as the first recommended solution. In response to finding the first candidate recommendation in the recommendation list, the first candidate recommendation is recommended to the user, and the first candidate recommendation is removed from the recommendation list.

2. The method according to claim 1, characterized in that, The method further includes: In response to the first candidate recommendation not being found in the recommendation list, a second recommendation for the user is obtained, wherein the second recommendation is of a different category than the first recommendation. Search the list of recommended solutions for a second candidate recommended solution that is of the same category as the second recommended solution. In response to finding the second candidate recommendation in the recommendation list, the second candidate recommendation is recommended to the user and then removed from the recommendation list.

3. The method according to claim 1, characterized in that, The method further includes: In response to the fact that no first candidate recommendation scheme is found in the recommendation scheme list and no candidate recommendation scheme is found in the recommendation scheme list, a second recommendation scheme list is generated, wherein the preset number ratios of the preset recommendation schemes of the M categories in the second recommendation scheme list are satisfied. Search the second recommended solution list for a third candidate recommended solution that is of the same category as the first recommended solution; In response to finding the third candidate recommendation scheme in the second recommendation scheme list, the third candidate recommendation scheme is recommended to the user, and the third candidate recommendation scheme is deleted from the second recommendation scheme list.

4. The method according to claim 1, characterized in that, The list of recommended solutions includes: Obtain the preset recommendation schemes for the M categories; The number N of candidate recommendation schemes is determined based on the preset balance time for the preset recommendation schemes for the M categories and the number of times the recommendation request is obtained within the preset balance time. The amplification factor is determined based on the number N of the candidate recommendation schemes and the preset quantity ratio; The preset recommended schemes for each category are amplified according to the amplification factor to form the recommended scheme list.

5. The method according to claim 1, characterized in that, The recommendation request includes the user's user characteristic information; The step of receiving a user's recommendation request and obtaining a first recommendation scheme for the user includes: The user's recommendation request is input into a preset recommendation model, so that the recommendation model determines multiple recommendation schemes for the user based on the user feature information in the recommendation request; The multiple recommended schemes are sorted from high to low according to their expected virtual value; The first recommendation solution is output from a plurality of recommendation solutions sorted from high to low based on the expected virtual value.

6. The method according to claim 5, characterized in that, The first recommended option is the one with the highest expected virtual value among the plurality of recommended options; The step of outputting the first recommendation scheme from a plurality of recommendation schemes sorted from high to low based on the expected virtual value includes: The recommendation scheme with the highest expected virtual value is selected from the multiple recommendation schemes and output as the first recommendation scheme.

7. A recommendation server, characterized in that, This is used to recommend products based on user characteristic information, including user personal information, user interests, user purchasing behavior, and user reading history. The recommendation server includes: a recommendation scheme list construction unit, a recommendation scheme acquisition unit, a candidate recommendation scheme query unit, and a recommendation unit. The recommendation scheme list construction unit is used to construct a recommendation scheme list, which includes N candidate recommendation schemes. Among the N candidate recommendation schemes, the preset recommendation schemes of M categories meet a preset quantity ratio. The N candidate recommendation schemes are obtained by expanding the preset recommendation schemes of the M categories according to the preset quantity ratio. The number N of candidate recommendation schemes is determined based on the preset balance time for the preset recommendation schemes of the M categories, and the historical acquisition number or expected acquisition number of recommendation requests within the preset balance time. The preset balance time is the unit time that represents the consumption of the candidate recommendation schemes in the recommendation scheme list. Where N and M are both positive integers greater than 1. The recommendation scheme acquisition unit is used to acquire a first recommendation scheme for the user in response to receiving a user's recommendation request; The candidate recommendation scheme query unit is used to query the recommendation scheme list for a first candidate recommendation scheme that is of the same category as the first recommendation scheme. The recommendation unit is configured to, in response to finding the first candidate recommendation scheme in the recommendation scheme list, recommend the first candidate recommendation scheme to the user and delete the first candidate recommendation scheme from the recommendation scheme list.

8. A recommendation system, characterized in that, The system includes: a user terminal, a business server terminal, and a recommendation server terminal as described in claim 7; wherein, The user terminal includes: a first receiving module, a first processing module, and a first sending module; The first receiving module is used to receive user operation instructions, wherein the operation instructions are actions that can trigger recommended behaviors; The first processing module is used to generate instruction information corresponding to the operation instruction based on the operation instruction; The first sending module is used to send the instruction information to the service server; The service server includes: a second receiving module, a second processing module, and a second sending module; The second receiving module is used to receive the instruction information sent by the user terminal; The second processing module is used to generate a recommendation request for the user based on the instruction information; The second sending module is used to send the recommendation request to the recommendation server; The recommendation server includes: a third receiving module; The third receiving module is used to receive the recommendation request sent by the business server.

9. An electronic device, characterized in that, include: Memory, processor; The memory is used to store one or more computer instructions; The processor is configured to execute one or more computer instructions to implement the method as described in any one of claims 1-6.

10. A computer-readable storage medium storing one or more computer instructions thereon, characterized in that, When this instruction is executed by the processor, it performs the method as described in any one of claims 1-6.

Images