CN111768239A

CN111768239A - Property recommendation method, device, system, server and storage medium

Info

Publication number: CN111768239A
Application number: CN202010608505.1A
Authority: CN
Inventors: 杜家春; 司雪敏
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-06-29
Filing date: 2020-06-29
Publication date: 2020-10-13

Abstract

The embodiment of the invention discloses a method, a device, a system, a server and a storage medium for recommending a road furniture; the embodiment of the invention can obtain and carry out feature extraction on the user information, the virtual prop information and the preset promotion scheme information to obtain the user features, M prop features and N promotion features; combining the property characteristics and the promotion characteristics to obtain MxN promotion property characteristics; combining the user characteristics and the promotion prop characteristics to obtain MxN fusion characteristics of the user; predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics; determining a target item in the virtual items based on the purchase probability, and determining a target promotion scheme in a preset promotion scheme; and recommending the target prop to the user by adopting a target promotion scheme. The embodiment of the invention customizes the promotion item for the user by comprehensively considering the possibility that the user purchases the virtual item under different promotion schemes. Therefore, the accuracy of the property recommendation method can be improved.

Description

Property recommendation method, device, system, server and storage medium

Technical Field

The invention relates to the field of computers, in particular to a method, a device, a system, a server and a storage medium for recommending a road tool.

Background

In the fields of e-commerce, virtual games, simulation experiments and the like, users can purchase virtual prop commodities by using virtual money, often recommend specific virtual prop commodities to the users in sales promotion activities, and provide certain sales promotion discounts to promote the consumption of the users.

At present, most of item recommendation methods classify users into different payment grades according to historical payment records of the users, and the lower the payment capacity of the users is, the cheaper the virtual item recommended to the users is, and the larger the sales promotion discount is.

However, the virtual item recommended by the method is often not purchased by the user because the recommended item is too expensive, the promotion discount is too small, the user does not need the virtual item, and the like, and thus, the accuracy of the current item recommendation method is low.

Disclosure of Invention

The embodiment of the invention provides a method, a device, a system, a server and a storage medium for recommending a property, which can improve the accuracy of the method for recommending the property.

The embodiment of the invention provides a method for recommending a piece of jewelry, which comprises the following steps:

acquiring user information, M pieces of virtual prop information and N pieces of preset promotion scheme information of a user, wherein N, M are positive integers which are greater than or equal to 1;

extracting the characteristics of the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics;

combining the M prop features and the N promotion features to obtain M multiplied by N promotion prop features, wherein the promotion prop features are characteristics of promotion props, and the promotion props are virtual props which are promoted by adopting a preset promotion scheme;

combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of the user purchasing the promotion props;

predicting the purchase probability of the user for purchasing the promotion item according to the fusion features;

determining target props in the M virtual props based on the purchase probability of the users for purchasing the promotion props, and determining a target promotion scheme in the N preset promotion schemes;

and recommending the target prop to the user by adopting the target promotion scheme.

The embodiment of the invention provides a road accessory recommendation device, which comprises:

the system comprises an acquisition unit, a processing unit and a display unit, wherein the acquisition unit is used for acquiring user information of a user, M pieces of virtual prop information and N pieces of preset promotion scheme information, and the N, M are positive integers which are more than or equal to 1;

the characteristic unit is used for carrying out characteristic extraction on the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics;

the combination unit is used for carrying out feature combination on the M prop features and the N promotion features to obtain M multiplied by N promotion prop features, wherein the promotion prop features are characteristics of promotion props, and the promotion props are virtual props which are promoted by adopting a preset promotion scheme;

the fusion unit is used for carrying out feature combination on the user features and the MXN promotion item features to obtain MXN fusion features of the user, and the fusion features are features of the user for purchasing the promotion items;

the prediction unit is used for predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics;

a determining unit, configured to determine a target item among the M virtual items based on a purchase probability of the user purchasing the promotional item, and determine a target promotional scheme among the N preset promotional schemes;

and the recommending unit is used for recommending the target prop to the user by adopting the target promotion scheme.

In some embodiments, the item information includes original price information of the virtual item, and the determining unit includes:

the expectation subunit is used for calculating the purchase expectation of the user for purchasing the promotion prop according to the original price information, the promotion information and the purchase probability;

a prop determining subunit, configured to determine, based on a purchase probability that the user purchases the promotional prop, a promotional prop that adopts the preset promotional scheme among the M virtual props;

the total expectation subunit is used for determining a total purchase expectation corresponding to the preset promotion scheme according to the purchase expectation of all promotion props adopting the preset promotion scheme;

the target scheme determining subunit is configured to determine a target promotion scheme among the N preset promotion schemes based on the total purchase expectation corresponding to each preset promotion scheme;

and the target prop determining subunit is used for determining the target props in the M virtual props.

In some embodiments, the desired subunit is to:

determining the price of the virtual prop after discount according to the original price information and the promotion information;

and calculating the purchase expectation of the user for purchasing the promotion prop according to the discounted price and the purchase probability.

In some embodiments, the prop determination subunit is to:

acquiring historical purchase quantity of the user;

determining a previous historical purchase quantity of the promotion items adopting the preset promotion scheme in the M virtual items according to the purchase probability of the user for purchasing the promotion items;

the total desired subunit to:

and accumulating the purchase expectation of the previous historical purchased number of the promotion items to obtain the purchase expectation of all the promotion items of the preset promotion scheme, and determining the total purchase expectation corresponding to the preset promotion scheme.

The total desired subunit further to:

acquiring user distribution information corresponding to the target promotion scheme;

sorting the users according to the total purchase expectation to obtain sorted users;

taking the users meeting the user distribution information from the sorted users as target users;

the recommending the target prop to the user by adopting the target promotion scheme comprises the following steps:

and recommending the target prop to the target user by adopting the target promotion scheme.

In some embodiments, the prediction unit comprises:

the acquisition subunit is used for acquiring a training sample set, wherein the training sample set comprises a plurality of training samples marked with real purchase marks of training users, the training samples comprise training fusion features, and the training fusion features are features of the training users for purchasing training promotion props;

the training subunit is used for training a preset model by adopting the training sample set until the preset model converges to obtain a prediction model;

and the predicting subunit is used for predicting the purchase probability of the user for purchasing the virtual item under the preset promotion scheme according to the fusion characteristics by adopting the prediction model.

In some embodiments, the obtaining subunit is configured to:

acquiring a real purchase label of a training user, training user information of the training user, training prop information and training promotion scheme information;

extracting features of the training user information, the training prop information and the training promotion scheme information to obtain training user features, training prop features and training promotion features;

combining the training prop characteristics and the training promotion characteristics to obtain training promotion prop characteristics, wherein the training promotion prop characteristics are characteristics of training promotion props, and the training promotion props are training props promoted by adopting a training promotion scheme;

performing feature combination on the training user features and the training promotion prop features to obtain training fusion features of the training user, so as to obtain a training sample, wherein the training fusion features are features of the training user purchasing the training promotion props;

and marking the training sample by adopting the real purchasing mark of the training user to obtain the training sample marked with the real purchasing mark of the training user.

In some embodiments, the obtaining subunit further includes:

an unpurchased submodule for determining the training samples labeled as unpurchased types of the true purchases as unpurchased training samples in the training sample set;

the screening submodule is used for screening the unpurchased training samples of the training users in the training sample set to obtain a screened training sample set;

at this point, in some embodiments, the training subunit is to:

and training a preset model by adopting the screened training sample set until the preset model is converged to obtain a prediction model.

In some embodiments, the screening submodule is to:

performing user clustering processing on the training users based on the training user information to obtain a plurality of training user clusters;

determining a training sample corresponding to each training user in the training sample set in the training user cluster;

and screening the training samples corresponding to the training sample set of each training user in the training user cluster to obtain the screened training sample set corresponding to the training user cluster.

In some embodiments, the preset model comprises a plurality of tree nodes, the training sample set comprises a plurality of the training samples, the training samples comprise real purchase labels, and the training submodule is configured to:

determining a historical tree node in the preset model, and acquiring node output of the historical tree node, wherein the historical tree node is a tree node before a current tree node;

determining a predicted gradient difference of the historical tree nodes based on the real purchase labels of the training samples and the node outputs of the historical tree nodes;

and modifying the parameters of the current tree node based on the prediction gradient difference of the historical tree nodes, determining the next tree node of the current tree node, returning to and executing the steps to determine the historical tree nodes in the preset model until the prediction gradient difference of the historical tree nodes is smaller than the preset gradient difference, and thus obtaining the prediction model.

The embodiment of the invention also provides a recommendation system, which comprises a master control end and a slave end, wherein,

the main control end is used for executing the following steps:

acquiring user characteristics and MxN promotion prop characteristics of a user, wherein N, M are positive integers which are greater than or equal to 1;

distributing the user characteristics of the user to subordinate terminals, and sending the promotion prop characteristics to all subordinate terminals;

receiving the target prop and the target promotion scheme of the user returned by the slave terminal so as to recommend the target prop to the user by adopting the target promotion scheme;

the slave is used for executing the following steps:

acquiring user characteristics and MxN promotional prop characteristics;

and returning the target prop and the target promotion scheme of the user to the master control end.

In some embodiments, when obtaining the user characteristics and the mxn promotional item characteristics, the slave is configured to:

acquiring user characteristics and a promotion prop characteristic table, wherein the promotion prop characteristic table comprises M multiplied by N promotion prop characteristics;

the performing feature combination on the user feature and the mxn promotional item features to obtain mxn fused features of the user, where the fused features are features of the user purchasing the promotional item, and the method includes:

and generating a user promotion prop characteristic table based on the user characteristics and the promotion prop characteristic table, wherein the user promotion prop characteristic table comprises M multiplied by N fusion characteristics.

The embodiment of the invention also provides a server, which comprises a memory, a storage and a control unit, wherein the memory stores a plurality of instructions; the processor loads instructions from the memory to execute the steps in any item recommendation method provided by the embodiment of the invention.

The embodiment of the present invention further provides a computer-readable storage medium, where multiple instructions are stored in the computer-readable storage medium, and the instructions are suitable for being loaded by a processor to perform steps in any item recommendation method provided in the embodiment of the present invention.

The embodiment of the invention can acquire user information of a user, M pieces of virtual prop information and N pieces of preset promotion scheme information, wherein N, M are positive integers which are more than or equal to 1; extracting the characteristics of the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics; combining the M prop features and the N promotion features to obtain M multiplied by N promotion prop features, wherein the promotion prop features are characteristics of promotion props, and the promotion props are virtual props which are promoted by adopting a preset promotion scheme; combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of the user purchasing the promotion props; predicting the purchase probability of the user for purchasing the promotion item according to the fusion features; determining target props in the M virtual props based on the purchase probability of the users for purchasing the promotion props, and determining a target promotion scheme in the N preset promotion schemes; and recommending the target prop to the user by adopting the target promotion scheme.

The method and the device predict the purchasing preference of the user to different virtual props under different promotion schemes by comprehensively considering the influence of the connection between the promotion scheme and the virtual props on the user during shopping, so that the virtual props which are most suitable for being recommended to the user are selected from the promotion schemes and the virtual props, the promotion scheme of the virtual props is customized for the user, the promoted virtual props are recommended to the user, the promoted virtual props have proper prices, and the user can be attracted to purchase. Therefore, the accuracy of the property recommendation method is improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1a is a scene schematic diagram of a prop recommendation method provided in an embodiment of the present invention;

FIG. 1b is a schematic flow chart of a method for recommending props according to an embodiment of the present invention;

fig. 2a is a schematic flow chart of a stage of item recommendation method applied to a training stage and a prediction stage in a game item recommendation scenario according to an embodiment of the present invention;

fig. 2b1 is a schematic diagram of a recommendation page of the item recommendation method provided by the embodiment of the present invention applied to a game item recommendation scenario;

fig. 2b2 is a schematic diagram of a recommendation page of the item recommendation method provided by the embodiment of the present invention applied to a game item recommendation scenario;

FIG. 2c is a schematic diagram comparing a first effect of the method for recommending props according to the embodiment of the present invention;

FIG. 2d is a schematic diagram comparing a second effect of the item recommendation method according to the embodiment of the present invention;

fig. 3 is a first structural schematic diagram of a prop recommendation device provided in the embodiment of the present invention;

fig. 4 is a schematic structural diagram of a server according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The embodiment of the invention provides a method, a device, a system, a server and a storage medium for recommending a piece of road furniture.

The property recommendation device may be specifically integrated in an electronic device, and the electronic device may be a terminal, a server, or other devices. The terminal can be a mobile phone, a tablet Computer, an intelligent bluetooth device, a notebook Computer, or a Personal Computer (PC), and the like; the server may be a single server or a server cluster composed of a plurality of servers.

In some embodiments, the item recommendation device may also be integrated into multiple electronic devices, for example, the item recommendation device may be integrated into multiple servers, and the multiple servers implement the item recommendation method of the present invention.

For example, in some embodiments, in order to improve recommendation efficiency, the item recommendation device may be specifically integrated in a server cluster, where the server cluster may include a master server and a slave server.

The property recommending device integrated in the master control end server can obtain user characteristics and MxN promotion property characteristics of a user, wherein N, M are positive integers which are more than or equal to 1; distributing the user characteristics of the user to the slave terminals, and sending the promotion prop characteristics to all the slave terminals; and receiving the target prop and the target promotion scheme of the user returned by the slave terminal so as to recommend the target prop to the user by adopting the target promotion scheme.

The property recommending device integrated in the slave server can obtain user characteristics and MxN promotional property characteristics; combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of the user purchasing the promotion props; predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics; determining a target item in the M virtual items based on the purchase probability of the user for purchasing the promotion item, and determining a target promotion scheme in the N preset promotion schemes; and returning the target prop and the target promotion scheme of the user to the master control end.

In some embodiments, in order to further improve recommendation efficiency, the master may send M × N promotional item features to the slave in a promotional item feature table, where the promotional item feature table includes M × N promotional item features, and the promotional item features may be formed by combining item features of virtual items and promotional features of a preset promotional scheme, where the promotional item features represent features of virtual items promoted by using the preset promotional scheme;

when the slave terminal acquires the user characteristics and the MxN promotion item characteristics, the slave terminal can acquire the user characteristics and the promotion item characteristic table, then the user promotion item characteristic table is generated based on the user characteristics and the promotion item characteristic table, the promotion item characteristic table comprises the MxN fusion characteristics, the purchase probability of the user for purchasing the promotion item is predicted by adopting the user promotion item characteristic table, therefore, the target item is determined in the virtual item based on the purchase probability of the user for purchasing the promotion item, the target promotion scheme is determined in the preset promotion scheme, and finally, the target item and the target promotion scheme of the user are returned to the master control terminal.

In some embodiments, the server may also be implemented in the form of a terminal.

For example, referring to fig. 1a, the electronic device may be a server, and the server may obtain user information of a plurality of users from a database, where the user information a of a user a is included, and obtain item information X of a virtual item X and item information Y of a virtual item Y, and promotion information of a preset promotion scheme 1, promotion information of a preset promotion scheme 2,. promotion information of a preset promotion scheme N, where N is a positive integer greater than or equal to 1.

Wherein, predetermine the promotion scheme and can be set up by technical staff, for example, set up 2 and predetermine the promotion scheme, be 9 promotion schemes and 7 promotion schemes of rolling over respectively, wherein, 9 promotion schemes of rolling over include 9 promotion information: the 0.9, 7-fold promotion scheme includes 7-fold promotion information: 0.7.

in some embodiments, the preset promotion scheme is a plurality of independent promotion schemes, and any virtual prop may select one promotion scheme from the preset promotion schemes to promote promotion, for example, the 9-fold promotion scheme and the 7-fold promotion scheme may be independent promotion schemes, and in a promotion scene, after the virtual prop X after 9-fold, the virtual prop X after 7-fold, the virtual prop Y after 9-fold, and the virtual prop Y after 7-fold may be provided after the virtual prop X and the virtual prop Y are promoted by using the two promotion schemes.

In some embodiments, a specific virtual item has a plurality of specific preset promotion schemes, for example, virtual item X has only a 9-fold promotion scheme, and virtual item Y has only a 7-fold promotion scheme, so that in a promotion scenario, after promotion, there may be a virtual item X after 9-fold and a virtual item Y after 7-fold.

The server can extract features of user information a, prop information x, prop information y, promotion information of 7 folds and promotion information of 9 folds, and arrange and combine the prop information and the promotion information respectively to obtain the fusion features of the user A, different virtual props and different promotion information, and the fusion features can embody the feature that the user purchases a certain virtual prop under a certain preset promotion scheme.

Specifically, the server may perform feature combination on the property features and the promotion features to obtain promotion property features, where the promotion property features are features of a virtual property promoted by using a preset promotion scheme, and then perform feature combination on the user features and the promotion property features to obtain fusion features, where the fusion features are features of a virtual property promoted by using a preset promotion scheme purchased by a user;

for example, in fig. 1a, feature extraction may be performed on user information a, prop information x, and 9-fold promotion information to obtain a fusion feature a _ x _ 9; extracting the characteristics of the user information a, the prop information x and the 7-fold promotion information to obtain a fusion characteristic a _ x _ 7; extracting the characteristics of the user information a, the prop information y and the 9-fold promotion information to obtain a fusion characteristic a _ y _ 9; and extracting the characteristics of the user information a, the prop information y and the 7-fold promotion information to obtain a fusion characteristic a _ y _ 7.

And then, predicting the purchase probability of the user for purchasing the virtual item under the corresponding preset promotion scheme according to the fusion characteristics.

For example, in fig. 1a, the probability P _ AX9 that user a purchases virtual item X with 9 folds can be predicted according to the fusion feature a _ X _ 9; predicting the probability P _ AY9 of the user A for purchasing the virtual item X of 9 folds according to the fusion feature a _ y _ 9; predicting the probability P _ AX7 of the user A for purchasing the 7-fold virtual item X according to the fusion feature a _ X _ 7; and predicting the probability P _ AY7 of the user A for purchasing the 7-fold virtual item Y according to the fused feature a _ Y _ 7.

Therefore, the target prop can be determined in the virtual props based on the purchase probabilities, and the target promotion scheme can be determined in the preset promotion scheme; for example, the purchase probabilities are ranked, the virtual item corresponding to the purchase probability with the highest value is determined as the target item, the promotion scheme with the highest value is determined as the target promotion scheme, and so on.

Finally, recommending target props to the user by adopting the target sales promotion scheme; for example, the target promotion scheme and the target prop are sent to the client of the user, the client of the user may display a recommendation page, the original price of the target prop and the discounted price after promotion according to the target promotion scheme may be displayed on the recommendation page, and the like.

The invention can consider the influence of the property promotion scheme on the property purchase of the player by considering different virtual properties and different promotion schemes as a whole in a characteristic combination mode, thereby knowing the property preference of the player and the price preference of different virtual properties, so as to recommend the property which is interested by the user at a proper price in a recommendation scene, avoid recommending the property which is interested by the user with a high price to the player without the payment capability, ensure that the attraction of the recommended property to the player is higher, and the price of the recommended property is in an acceptable range of the player.

The following are detailed below. The numbers in the following examples are not intended to limit the order of preference of the examples.

Artificial Intelligence (AI) is a technique that uses a digital computer to simulate the human perception environment, acquire knowledge, and use the knowledge, which can make a machine function similar to human perception, reasoning, and decision making. The artificial intelligence technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning, deep learning and the like.

Among them, Machine Learning (ML) is a technology that uses a computer to replace a human brain to understand, learn, and further process a target, thereby making a computer realize intelligence. The machine learning techniques generally include deep learning, reinforcement learning, migration learning, teaching learning, inductive learning, transduction reasoning, analogy learning, deductive learning, game theory, and the like, and for example, the deep learning techniques may include the technical fields of artificial neural networks, attention learning, knowledge characterization, and the like.

In this embodiment, a prop recommendation method based on machine learning is provided, as shown in fig. 1b, a specific process of the prop recommendation method may be as follows:

101. and acquiring user information, M pieces of virtual prop information and N pieces of preset promotion scheme information of the user.

Wherein N, M are all positive integers greater than or equal to 1.

The user information of the user may include consumption information, personal information, log information, and the like of the user, among others. For example, personal information of a game player may include age information, gender information, etc. of the player; the consumption information can comprise the consumption information of the game appearance (such as game objects with character decorative effects, such as game skin, character head ornaments, weapon special effects) of the game player in the last month, the purchase quantity information of the game objects and the like; the log information refers to a series of user history information left by the game behavior of the user stored in the game log database, for example, the log information may include: the average daily online time of the game player, game winning rate information, player character level information, and the like.

Virtual items (PROPS) may broadly refer to any virtual object in a virtual scene that may be used to decorate arrangements, modify values, or provide enhancements, etc. For example, game items may include expendable items, equipment items, task items, character items, appearance items, and so forth. In some electronic games, game items may be obtained by players through the consumption of virtual or real currency purchases.

The property information of the virtual property (i.e. virtual property information, which will be referred to as virtual property information hereinafter) may include original price information, property type information, property attribute information, and the like of the virtual property.

The item type information is information with a type of the item, for example, the item type may include an appearance decoration type, a game role type, a social tool type, and the like.

The property information of the property is the information of the game property of the property, for example, the game property of the game property "red liquid medicine" is: increase 350 vital value within 10 seconds; for another example, the game attributes of the game prop "belt" are: the upper limit of the life value is increased by 500.

In some games, players may purchase different game characters (or heros, drymen, etc.) for game play, and different game characters may have different prices.

For example, some games may sell a game character item for a player to unlock the game character, thereby enabling the player to use in a game play, and the item information for the game character item may include character type information, character release time information, character attribute information, character price information, and the like for the game character item.

The item information of the game appearance item may include appearance type information, appearance grade information, appearance price information, appearance image information, and the like of the game appearance item.

For example, the information of the appearance type of the appearance prop may include a fine type, a unique type, a general type, and the like.

The preset sales promotion scheme is a marketing scheme which is made according to market demands in order to attract users to consume and expand sales. The marketing plan may have a plurality of preferential levels, for example, the preset promotion plan may include a high discount promotion plan, a medium discount promotion plan, a low discount promotion plan, a no discount promotion plan, etc. which are set by the developer according to the needs, for example, the high discount plan may be buy 1, the medium discount plan may be 7 discounts, the low discount plan may be full of 50 yuan for giving 10 diamonds, etc.; the predetermined promotion scheme may have a variety of scheme manifestations, for example, the predetermined promotion scheme may be a discount-based promotion scheme, a buy-away-based promotion scheme, a discount-based promotion scheme, a coupon-based promotion scheme, or the like.

The promotion information of the preset promotion scheme (i.e., the preset promotion scheme information, which will also be referred to as promotion scheme information, promotion information hereinafter) may be promotion value information, advertisement image information of the advertisement for promotion, advertisement text information, etc., and for example, in some embodiments, the preset promotion scheme may include 7-fold promotion, 8-fold promotion, and 9-fold promotion, which correspond to promotion information of 0.7, 0.8, and 0.9, respectively.

Further, in some embodiments, the pre-set promotional program can include buying 3 a-for-1 promotion, buying 2 a-for-1 promotion; in some embodiments, the predetermined promotion scheme may include a promotion of virtual item x at 100, a promotion of virtual item y at 120, and so on.

The method for acquiring the user information of the user, the property information of the virtual property, and the promotion information of the preset promotion scheme includes various methods, for example, acquiring the user information, the property information of the virtual property, and the promotion information of the preset promotion scheme from a server through a network, for example, inputting the user information into a local after being collected by a developer, reading the user information from a local memory, and the like.

It should be noted that the property information of the virtual property and the promotion information of the preset promotion scheme are often pre-established and rarely modified, so the updating frequency of the property information and the promotion information is low, and the user information (e.g. the user consumption information) of the user changes with the change of the user behavior each time, so the updating frequency of the user information is high. In some embodiments, in order to reduce the computational consumption and reduce the occupied network bandwidth, whenever it is necessary to acquire user information of a user, prop information of a virtual prop, and promotion information of a preset promotion scheme, the prop information of the virtual prop and the promotion information of the preset promotion scheme may not be acquired any more, but the last acquired prop information of the virtual prop and promotion information of the preset promotion scheme are adopted, and when the user information of the user is acquired, the latest user information may be acquired.

102. And carrying out feature extraction on the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user features, M pieces of prop features and N pieces of promotion features. The user information, the property information and the promotion information can be composed of character data, digital data, character symbol data, image data, audio data, video data and the like, for example, the user information can include a user name, a user head portrait, user game voice, user screen recording data and the like; the property information can comprise a property name, a property UI image, a property trigger sound effect, a property UI video and the like; the promotion information can comprise promotion characters, promotion UI images, promotion special effect audios, promotion special effect videos and the like; for example, in a game mall promotional scenario, when some game props are discounted, their prop UI pictures can be augmented with special audio and special video of a dynamic promotional effect, such as a flashing motion picture of a "7" word accompanied by a cheering sound.

The feature extraction method may adopt one-hot code (one-hot code) encoding, text embedding (word), Bag of Words (Bag of Words) model, voiceprint extraction, image edge feature extraction, image feature extraction based on machine learning, and the like.

Wherein, can adopt the one-hot code to encode to fixed quantity's data, the one-hot code can be according to N bit state register to encode N states, for example, prop UI image can adopt the one-hot code to encode: assuming that the game has 5 kinds of props, namely props A, B, C, D, E, each of which corresponds to a unique prop UI image, and is marked as images a, b, c, d, and e, the unique hot code of the prop UI image is (1, 0, 0, 0, 0) for prop a; for the property B, the one-hot code of the property UI image is (0, 1, 0, 0, 0); for prop C, the one-hot code of the prop UI image is (0, 0, 1, 0, 0), and for prop D, the one-hot code of the prop UI image is (0, 0, 0, 1, 0); for prop E, the one-hot code of the prop UI image is (0, 0, 0, 0, 1).

It should be noted that, since the user usually considers the promotion price of the virtual item when purchasing the virtual item, different discount prices often affect the purchase desire of the user for the same virtual item.

Therefore, in this embodiment, it is necessary to combine the user characteristics, item characteristics, and promotion characteristics to finally obtain a fusion characteristic such as [ player characteristics, (promotion scheme characteristics, item characteristics) ], so as to predict the possibility that a certain player purchases a certain virtual item under a certain promotion scheme, as shown in step 103 and step 104:

103. and combining the M prop features and the N promotion features to obtain M multiplied by N promotion prop features, wherein the promotion prop features are the features of promotion props, and the promotion props are virtual props which are promoted by adopting a preset promotion scheme.

For example, referring to fig. 1a, in step 103, each prop feature and each promotion feature may be cross-spliced two by two to obtain a cross-promotion prop feature corresponding to the prop feature and the promotion feature.

The promotion item features can simultaneously represent the promotion scheme and the features of the virtual items, namely, the promotion item features can be regarded as the features exhibited by the virtual items after promotion by adopting the promotion scheme.

For example, in some embodiments, item feature x, item feature y, promotion feature m, and promotion feature n are extracted, so a promotion item feature x _ m corresponding to item feature x and promotion feature m, a promotion item feature x _ n corresponding to item feature x and promotion feature n, a promotion item feature y _ m corresponding to item feature y and promotion feature m, and a promotion item feature y _ n corresponding to item feature y and promotion feature n can be obtained, that is, when considering the influence on the user's purchase, the same virtual item at multiple different promotions can be regarded as multiple different virtual items, for example, virtual item a with 7 folds and 9 folds is regarded as two different virtual items: virtual prop a is folded by 7, and virtual prop b is folded by 9.

104. And combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of the user purchasing the promotion props.

Then, the user characteristic of the user and the promotion item characteristic are combined in a pairwise manner to obtain N fusion characteristics, such as [ player characteristic, (promotion scheme characteristic, item characteristic) ].

The feature combination mode can be feature splicing, feature accumulation, feature multiplication and the like; for example, for the feature [ a ] and the feature [ B ], the features [ a, B ] may be obtained by concatenation, the features [ a + B ] may be obtained by accumulation, the features [ a × B ] may be obtained by multiplication, and the like.

105. And predicting the purchase probability of the user for purchasing the promotion prop according to the fusion characteristics.

The prediction method can adopt a machine learning method to predict.

For example, in some embodiments, a machine learning method is used to predict the purchase probability of the user purchasing the virtual item under the preset promotion scheme according to the fusion features, so that the item recommendation method can be more accurate, and step 103 may include the following steps:

(1) acquiring a training sample set, wherein the training sample set comprises a plurality of training samples marked with real purchase marks of training users, the training samples comprise training fusion characteristics, and the training fusion characteristics are characteristics of the training users for purchasing training promotion props;

(2) training a preset model by adopting a training sample set until the preset model converges to obtain a prediction model;

(3) and predicting the purchase probability of the user for purchasing the virtual item under the preset promotion scheme by adopting the prediction model according to the fusion characteristics.

The training samples may be obtained from a database through a network, or may be imported into a local memory by a technician and read from the local memory, and so on.

The preset model can be any one of prediction models such as a neural network model and a traditional machine learning model.

The preset model may adopt a Gradient Boosting iterative decision tree (GBDT) algorithm, or may also adopt an XGBoost (enhanced Gradient Boosting) algorithm for training, and so on.

The GBDT generates a weak classifier through multiple iterations, each iteration generates one weak classifier, and each weak classifier is trained on the basis of the residual error of the last iteration, so that the precision of the final classifier is continuously improved by reducing the deviation of the weak classifier in the training process.

The weak classifier is a classifier whose Classification result is slightly better than that of the random prediction, but the accuracy is not too high, And may be, for example, a Classification And Regression Tree (CART).

The step of obtaining the training sample may include the following steps:

the training and promotion property characteristics are combined with the training and promotion characteristics to obtain training and promotion property characteristics, the training and promotion property characteristics are the characteristics of the training and promotion property, and the training and promotion property is a training property promoted by adopting a training and promotion scheme;

the training user characteristics and the training promotion property characteristics are combined to obtain training fusion characteristics of the training user, so that a training sample is obtained, and the training fusion characteristics are characteristics of the training user for purchasing the training promotion property;

and marking the training samples by adopting the real purchasing marks of the training users to obtain the training samples marked with the real purchasing marks of the training users.

In some embodiments, the real purchase label may be set to 1 for training samples that train users to actually purchase training virtual props of the training promotion scheme, and the real purchase label may be set to 0 for training samples that train users not to purchase training samples despite the training virtual props for which the training promotion scheme is recommended.

In some embodiments, the step of "obtaining training samples" may comprise the steps of:

A. determining training samples marked as unpurchased types of real purchases as unpurchased training samples in the training sample set;

B. screening unpurchased training samples of training users in the training sample set to obtain a screened training sample set;

therefore, in some embodiments, the step of "training the predetermined model using the training sample set until the predetermined model converges to obtain the prediction model" may include the following steps:

C. and training the preset model by adopting the screened training sample set until the preset model is converged to obtain a prediction model.

Since the number of people who purchase the promotional item is necessarily smaller than the number of people who are recommended with the promotional item but who do not purchase the promotional item, in order to ensure the balance of the training sample, the screening process refers to a series of data preprocessing operations such as selection and deletion of sample individuals, for example, the screening process may include down-sampling, such as hierarchical down-sampling. In some embodiments, a training sample labeled 0 may be downsampled for a true purchase.

The down-sampling refers to a given data set S, and the data set S is processed by an acquisition function to generate a subset S ', so that the spatial resolution of S is higher than that of S'.

In some embodiments, the training sample with the true purchase label of 0 may be downsampled such that the spatial resolution of the training sample with the true purchase label of 0 after downsampling is lower than the spatial resolution of the training sample with the true purchase label of 0 before downsampling; in some embodiments, the spatial resolution of the training samples labeled 0 after down-sampling can also be made the same as or similar to the spatial resolution of the training samples labeled 1 for true purchase.

Because the user information caused by the change of the user behavior is dynamically changed, and the property information is basically stable and unchanged, in some embodiments, in order to reduce the calculated amount and reduce the network occupancy rate, the training property information and the training promotion information in the training sample acquired when the preset model is trained can be the same as the property information and the promotion information acquired when the prediction model is adopted, that is, when the prediction model is adopted for prediction, the training property information and the training promotion information adopted when the model is trained last time can be adopted.

In some embodiments, the user information of the current day of the training process may be obtained from the database as the training user information for the training process, and the user information of the current day of the prediction process may be obtained from the database for the prediction process.

In some embodiments, when hierarchical down-sampling is adopted, the step "b. performing a screening process on unpurchased training samples of a training sample set for training a user, to obtain a screened training sample set" may include the following steps:

b₁performing user clustering processing on the training users based on the training user information to obtain a plurality of training user clusters;

b₂determining a training sample corresponding to each training user in the training sample set in the training user cluster;

b₃and screening the training sample corresponding to each training user in the training user cluster in the training sample set to obtain the screened training sample set corresponding to the training user cluster.

The training samples can be divided into a plurality of clusters according to information with classification properties in the training samples, such as user payment information and the like, and then a part of the training samples in each cluster are selected as the screened training samples, so that the screened training samples can cover various types of users.

In some embodiments, the preset model may include a plurality of tree models (i.e., weak classifiers), the training sample set may include a plurality of training samples, the training samples may include real purchase labels, and the step "c. training the preset model using the training sample set until the preset model converges to obtain the prediction model" may include the steps of:

determining historical tree nodes in a preset model, and acquiring node output of the historical tree nodes, wherein the historical tree nodes are tree nodes before the current tree node;

determining a prediction gradient difference of the historical tree nodes based on the real purchase labels of the training samples and the node output of the historical tree nodes;

In some embodiments, the preset model may be trained by using a GBDT algorithm, where the GBDT algorithm is obtained by weighted summation of weak classifiers obtained from each training cycle, for example, the preset model may be obtained by weighted summation of classification regression trees obtained from each training cycle, and the preset model may be finally described as:

wherein M is the iteration number of model training, and each round generates a classification regression tree T (x; theta)_m) To optimize the loss function

The following were used:

where the current training is iterated i times, f_m-1(x_i) For the preset model of the last iteration, y_iFor output of a predetermined model, θ_mFor the loss function of the last weak classifier, L may be a square loss function, and by continuous iterative training, the GBDT may reduce the loss function along the direction of gradient descent, so as to fit a negative gradient of the loss function under the current model in each iteration, so that the loss function may be continuously reduced as soon as possible, and the accuracy of the training result is ensured.

Wherein the square loss function is as follows:

L(y_i，f(x_i))＝(y_i-f(x_i))²

therefore, the flow of the loss function algorithm of the GBDT is as follows:

1. initialization f₀(x) The following were used:

wherein c is a preset initial parameter;

2. at the mth iteration, the residual r may be calculated_miThe following were used:

from the residual error r_miFitting a classification regression tree, and then obtaining the output value c of the leaf node j in the classification regression tree by minimizing the current loss function_mj:

c_mi＝argmin∑L(y_i,f_m-1(x_i)+c)

Updating the preset model as follows:

in some embodiments, the preset model may be trained using the GBDT algorithm, wherein each time a tree model is added (i.e., iterated once), the difference between the sum of all tree models in the preset model of the previous iteration and the true gradient may be fitted.

The XGboost is similar to the GBDT, and the difference is that the XGboost adds a regularization term to a loss function, and the loss function J of the XGboost is as follows:

J(f(x_i))＝L(f(x_i))+Ω(f(x_i))

compared to the loss function of GBDT, a regularization term Ω is also added, where:

compared with GBDT which takes CART as a weak classifier, XGboost can also adopt a linear classifier (Linear classifier) as the weak classifier, and a regular term omega is added in XGboost, which can be used for controlling the complexity and the balance of a model, so that a learned prediction model is simpler, and the problem of overfitting is prevented.

106. And determining a target item in the M virtual items based on the purchase probability of the user for purchasing the promotion item, and determining a target promotion scheme in the N preset promotion schemes.

Step 105 may predict a purchase probability of the user to purchase the virtual item under the preset promotion scheme, where the higher the purchase probability, the more likely the user is to purchase the promotion item. The promotion item refers to a virtual item recommended under a preset promotion scheme.

In some embodiments, the preset promotion scheme corresponding to the promotion item with the highest purchase probability may be directly determined as the target promotion scheme, and the virtual item corresponding to the promotion item with the highest purchase probability may be determined as the target item.

In some embodiments, a preset promotion scheme corresponding to the first N promotion items with the highest purchase probability may be directly determined as a target promotion scheme, and a target item may be determined according to the historical purchased items of the user, where N may be a positive integer set according to market requirements.

It should be noted that the target prop may be one or more, and the target promotion scheme may also be one or more.

In addition, the too much promotion may result in a situation that the purchase rate is high but the total benefit is high, and for the game property, the too much promotion may also result in the property devaluation problem in the game, so in some embodiments, in order to solve the above problem, a Greedy Algorithm (Greedy Algorithm) may be used to determine the target promotion scheme, where the property information may include the original price information of the virtual property, and step 104 may include the following steps:

the method comprises the following steps that the target prop is determined in M virtual props according to the purchase probability, and the target promotion scheme is determined in N preset promotion schemes, and comprises the following steps:

(1) calculating the purchase expectation of the user for purchasing the promotion props according to the original price information, the promotion information and the purchase probability;

(2) determining a promotion item adopting a preset promotion scheme in the M virtual items based on the purchase probability of the user for purchasing the promotion item;

(3) determining the total purchase expectation of the target props corresponding to the preset promotion scheme according to the purchase expectation of all promotion props adopting the preset promotion scheme;

(4) determining a target promotion scheme in the N preset promotion schemes based on the total purchase expectation corresponding to each preset promotion scheme;

(5) and determining a target prop from the M virtual props.

In some embodiments, the step (1) "calculating the purchase expectation of the user to purchase the promotional item according to the original price information, the promotion information, and the purchase probability" may include the steps of:

The original price information may include an original price value I of the virtual item I, the promotion information may include a discount value β of a preset promotion scheme B, and the purchase expectation E of the user a for the virtual item I under the preset promotion scheme B_AThe calculation formula of (a) is as follows:

E_A＝β*i*P_ABI

wherein, P_ABIAnd the purchase probability of the user A for purchasing the virtual prop I under the preset promotion scheme B.

The method includes the steps that M target props under a preset promotion scheme can be determined in a plurality of virtual props based on purchase probability, wherein M can be historical purchase quantity, in some embodiments, M can be average quantity of props purchased by a user at a historical moment, and user information can include information of the historical purchase quantity.

Thus, in some embodiments, the step "(2) of determining, among the M virtual items, a promotional item that employs a preset promotional scheme based on a purchase probability of the user purchasing the promotional item" may include the steps of:

acquiring historical purchase quantity K of a user;

and determining the first K target promotion props adopting a preset promotion scheme in the M virtual props according to the purchase probability of the users for purchasing the promotion props. At this time, the step "(3) of determining the total purchase expectation corresponding to the preset promotion scheme according to the purchase expectation of all the promotion items adopting the preset promotion scheme" includes:

and accumulating the purchase expectation of the target props which are historically purchased in quantity and promoted under each preset promotion scheme to obtain the purchase expectation of all promotion props of the preset promotion scheme and determine the purchase total expectation corresponding to the preset promotion scheme.

For example, in some embodiments, the purchase probabilities of the user a purchasing each virtual item under the preset promotion scheme B may be ranked, then the number M of the users a historically purchasing the average virtual items under the preset promotion scheme B is determined, and the top M virtual items are determined as the target items in the virtual items ranked from high to low according to the purchase probabilities.

For example, if it is determined by the game server that the number of game items purchased by the user in a week is 12, the top 12 promotion items using the preset promotion scheme in the M sorted virtual items may be determined as the target promotion items in the promotion items purchased by the user, sorted from high to low according to the purchase probability.

The target promotion scheme may be determined in the preset promotion scheme based on a total purchase expectation of the user for all target props under the preset promotion scheme.

For example, in some embodiments, the step "(3) of determining the total purchase expectation of all target items promoted by the preset promotion scheme according to the purchase expectation of the target items" may include the following steps:

and accumulating the purchase expectation of the first M target props which are promoted by adopting the preset promotion scheme to obtain the purchase total expectation of all the target props under the preset promotion scheme.

For example, in some embodiments, the step "(2) after determining the target promotion program" in the preset promotion program based on the total purchase expectation of the user for all the target props under the preset promotion program, may further include the following steps:

sequencing the users according to the total purchase expectation to obtain the sequenced users;

taking the users which accord with the user distribution information from the sorted users as target users;

at this time, the step "(5) of recommending the target item to the user by using the target promotion scheme" may be recommending the target item to the target user by using the target promotion scheme.

For example, the total purchase expectation of the user a for all the target items under the preset sales promotion scheme B is the sum of the purchase expectation of the user a for each target item under the preset sales promotion scheme B, and after sorting is performed according to the magnitude of the total purchase expectation, the preset sales promotion scheme corresponding to the total purchase expectation with the largest value may be determined as the target sales promotion scheme.

The user distribution information may refer to allocating different promotion schemes to different user groups according to market demands, for example, the user distribution information may be [ 30% of players before the strength chart enjoy 9-fold benefits, 70% of players before the strength chart enjoy 7-fold benefits ], and for example, the user distribution information may be [ 2000 players before the strength chart enjoy one-out-of-business benefits, 500 players before the strength chart enjoy 5-fold benefits, 100 players before the strength chart enjoy 1-fold benefits ], and so on.

107. And recommending the target prop to the user by adopting a target promotion scheme.

The method for recommending the target prop to the user by adopting the target promotion scheme has various modes, for example, the target promotion scheme and the target prop are directly sent to the user; for example, a promotion image is generated according to the target promotion scheme and the target prop, and the promotion image is sent to the user, and the like.

In some embodiments, the target promotion scheme and the target prop may be directly sent to a client of the user, and the client generates a promotion page of a mall according to the target promotion scheme and the target prop, so as to display the target prop promoted by using the target promotion scheme to the user.

The recommendation scheme provided by the embodiment of the invention can be applied to various recommendation scenes, for example, by taking game item recommendation as an example, the scheme can be used for recommending interesting game items to players in a personalized manner according to consumption habits and game habits of different players and adopting a proper promotion discount to attract the players to purchase. By adopting the scheme provided by the embodiment of the invention, the promotion scheme accepted by the player can be predicted more accurately, the price of the property after promotion is not too low, and the game property which is most likely to be purchased can be recommended to the player by the scheme, so that the recommendation precision is further improved.

Therefore, the embodiment of the invention can acquire the user information and the virtual prop information of the user and preset the promotion scheme information; extracting characteristics of the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics; combining the M prop features and the N promotion features to obtain M multiplied by N promotion prop features, wherein the promotion prop features are characteristics of promotion props, and the promotion props are virtual props which are promoted by adopting a preset promotion scheme; combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of the user purchasing the promotion props; predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics; determining a target item in the M virtual items based on the purchase probability of the user for purchasing the promotion item, and determining a target promotion scheme in the N preset promotion schemes; and recommending the target prop to the user by adopting a target promotion scheme.

Therefore, according to the scheme, the personalized property and promotion customization of the user can be carried out by adopting a machine learning method and combining the user information, the property information and the promotion information, the property which is interested in each user and the receivable promotion price are provided for each user, the promotion scheme and the influence of the property on the shopping of the user are considered, and the accuracy of the property recommendation method is improved.

The method described in the above embodiments is further described in detail below.

In this embodiment, the method of the embodiment of the present invention will be described in detail by taking the case of recommending a game item to a player in an electronic game as an example.

As shown in fig. 2a, in order to solve the above problem, the present embodiment provides a method for recommending props, which includes the following specific processes:

201. a training sample set and a prediction sample set are obtained from a game database.

In order to ensure that the characteristics of the game item and the promotion scheme are considered when the game item is recommended to the player, a certain game item adopting a certain promotion scheme can be considered as a new game item.

In this embodiment, the game item may be a game character item, a character appearance item, or the like.

The training sample set can comprise a plurality of training samples, and each training sample is composed of player characteristics of a player, prop characteristics of a game prop, scheme characteristics of a promotion scheme and real purchase marks.

Wherein, the real purchasing label can be manually labeled or obtained from a game database; when the true purchase mark is 1, it means that the player in the training sample purchased the game item under the promotion scheme, and when the true purchase mark is 0, it means that although the game item under the promotion scheme is recommended to the player in the training sample, the player did not purchase.

Considering that there is a natural quantity difference between the number of purchasers and the number of unpurciers, the training sample labeled 0 may be down-sampled in this embodiment to ensure sample balance and improve recommendation accuracy.

Similarly, the prediction sample set may include a plurality of prediction samples, each of which is composed of player characteristics of a player, item characteristics of a play item, and scheme characteristics of a promotion scheme.

The characteristics of the player can be obtained by extracting the characteristics in the player information, the characteristics of the prop can be obtained by extracting the characteristics in the prop information, and the characteristics of the promotion scheme can be obtained by extracting the characteristics in the promotion scheme information; the player information, the item information, and the promotional program information can all be from a game database.

The player characteristics may include a pay characteristic that represents the player's ability to pay, and an active characteristic that represents the player's preference for game items, among other things.

The payout characteristics may include, among other things, the player's average ticket consumption, the game character purchase amount, the character appearance purchase amount, the draw versus directly purchased ticket consumption, and so on.

The active features may include, among other things, the average online time of the player, the number of competitions, the winning rate, the number of game characters the player has owned, the number of appearances of the player's owned characters, the level and position of the player, etc.

The game props can be divided into game role props, role appearance props and the like. The prop characteristics of the game role prop can comprise an attribute value of the game role, occupation of the game role, original price of the game role, shelf life of the game role and the like; the prop characteristics of the character appearance prop can comprise a character appearance type, a character appearance grade, a purchasing mode of the character appearance, an original price of the character appearance, a picture characteristic of the character appearance and the like.

In this embodiment, the game items and promotional programs are substantially stable and do not change over time, so the same set of item features and promotional program features can be used when obtaining the training sample set and the prediction sample set from the game database.

And because the characteristics of the players change along with the behaviors of the players, the latest user information in the game database is extracted aiming at the characteristics of the players whenever the training sample set and the prediction sample set are acquired, so that the recommendation accuracy is ensured.

202. And training the preset model by adopting a training sample set until the preset model is converged to obtain a prediction model.

In this embodiment, the XGBoost algorithm may be adopted to train the preset model to obtain the prediction model.

For a specific training process, reference may be made to step 105, which is not described herein in detail.

203. And predicting the prediction sample set by adopting a prediction model, and determining the purchase probability of each player purchasing different game props in the prediction sample set under the promotion scheme.

Assuming that the number of property features is i, the number of promotion features is j, and the number of users is k, after pairwise cross-splicing processing is performed on each property feature and each promotion feature in step 103, i × j promotion property features will be obtained, and in step 104, the number of fusion features of all users is required to be obtained as i × j k.

Therefore, the recommendation scene has a strict requirement on the effectiveness, and when the number of the user features, the prop features and the promotion scheme features is too large, the recommendation effect is often not expected.

Therefore, in some embodiments, a recommendation system adopting a Master-Slave Model (Master-Slave Model) may be designed to extract i × j × k fusion features according to the prop information of i virtual props, the promotion information of j promotions, and the user information of k users, so as to solve the problem of low efficiency caused by large-scale data processing.

Specifically, in this embodiment, a recommendation system with a master-slave architecture is provided, so as to perform the steps of "obtaining user information of a user, prop information of M virtual props, and N pieces of preset promotion scheme information, where each promotion information, N, M is a positive integer greater than or equal to 1; extracting characteristics of the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme promotion information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics; combining the M prop features and the N promotion features to obtain M multiplied by N promotion prop features, wherein the promotion prop features are the features of promotion props, and the promotion props are the features of virtual props which are promoted by adopting a preset promotion scheme; combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of virtual props which are promoted by the user by adopting a preset promotion scheme when the user buys the promotion props; predicting the purchase probability of the user for purchasing the virtual promotion item under the condition of purchasing a preset promotion scheme according to the fusion characteristics; determining a target item in the M virtual items based on the purchase probability of the user to purchase the promotion item, and determining a target promotion scheme' in the N preset promotion schemes:

the recommendation system may include a Master and a slave, wherein the Master (Master) is configured to perform the following steps:

acquiring user characteristics of a user and characteristics of M multiplied by N promotion props, wherein N, M are positive integers which are more than or equal to 1;

distributing the user characteristics of the user to the slave terminals, and sending the promotion prop characteristics to all the slave terminals;

and the Slave (Slave) is configured to perform the following steps:

acquiring user characteristics and MxN promotional prop characteristics;

predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics;

determining a target item in the M virtual items based on the purchase probability of the user for purchasing the promotion item, and determining a target promotion scheme in the N preset promotion schemes;

The Master-Slave architecture may include a Master (Master) and a plurality of slaves (Slave).

The master control end can extract user information of the user, prop information of the M virtual props and N pieces of preset promotion scheme information into user characteristics and the MXN promotion prop characteristics of the user by a characteristic extraction method.

For a specific feature extraction method, reference may be made to steps 102 and 103, which are not described herein again.

In order to consider the influence of the promotion scheme and the game props on the purchase desire of the player, in some embodiments, the master control end may splice the promotion scheme features into the training sample together as a part of the prop features, and perform unique hot coding on the spliced features.

In some embodiments, the property features and the promotion scheme features can be combined in pairs in the slave end, namely, game properties with different discounts are regarded as different game properties, and the game properties have different property-discount features.

For example, in some embodiments where the promotional program includes 3, five-fold, six-fold, and seven-fold promotions, the feature vector of the promotional program feature may be determined to be a three-dimensional vector, as shown in Table 1:

	five-fold promotion	Six-fold promotion	Seven-fold promotion
				Feature vector for 5-fold promotion	1	0	0
Feature vector for 6-fold promotion	0	1	0
				Features for 7-fold promotionVector quantity	0	0	1

TABLE 1

The position of the sales promotion scheme corresponding to the discount can be 1, and the other positions can be 0, for example, the feature vector of the sales promotion with 5 folds is (1, 0, 0), the feature vector of the sales promotion with 6 folds is (0, 1, 0), and the feature vector of the sales promotion with 7 folds is (0, 0, 1).

And finally, the subordinate terminal arranges and combines the promotion scheme characteristics, the prop characteristics and the player characteristics, namely splices the promotion scheme characteristics, the prop characteristics and the player characteristics into fusion characteristics [ the player characteristics, the prop characteristics and the promotion scheme characteristics ].

The order of magnitude of players in a game scene can reach hundreds of millions, the number of game props can be more than hundreds, and promotion schemes can be various, because the characteristics of the players, the characteristics of the props and the characteristics of the promotion schemes are arranged and combined, the number of the obtained fusion characteristics is too large, and a large amount of computing resources and computing time are consumed, so that the purchasing probability can be predicted by adopting a master-slave mode of a Spark platform (a distributed open-source cluster operation framework) in a model prediction stage.

The Spark platform comprises a master control end and a plurality of slave ends, the master control end can distribute user characteristics and promotion item characteristics of a player to the slave ends, the slave ends can combine the user characteristics and the promotion item characteristics to obtain fusion characteristics of the user, a prediction model is adopted to predict the purchase probability of the player for purchasing the virtual item under a preset promotion scheme according to the fusion characteristics, the target item is determined in the virtual item based on the purchase probability of the user for purchasing the promotion item, and the target promotion scheme is determined in the preset promotion scheme.

The master control end of the Spark platform comprises a Cluster management server (Cluster Manager), and the Cluster Manager can control and monitor all slave ends;

the slave of Spark platform includes a Worker node server (Worker) that can control computing resources, such as a launch Executor (Executor) to run a process on the Worker.

In some embodiments, in order to reduce the problem of data expansion, after extracting the user characteristics and the promotional item characteristics, the master may assign one user characteristic to each slave, and broadcast a promotional item characteristic table to all slaves, where the promotional item characteristic table includes M × N promotional item characteristics.

After the subordinate terminal obtains the user characteristic and the promotion item characteristic table, a user promotion item characteristic table can be generated based on the user characteristic and the promotion item characteristic table, and the user promotion item characteristic table comprises M × N fusion characteristics.

For example, the master may broadcast the promotional item feature table X to all slaves, and send the user feature a of user a to slave 1, the user feature B of user B to slave 2, and the user feature C of user C to slave 3.

For the slave end 2, the slave end 2 can combine the user characteristics and the promotion item characteristic table to generate a user promotion item characteristic table or a matrix; for example, the user characteristic is added to each unit of the promotion item characteristic table, and finally, a user promotion item characteristic table containing M × N user promotion item characteristics is obtained, where the user promotion item characteristics are fusion characteristics.

In order to further improve the efficiency of feature fusion and prevent data expansion, in some embodiments, the master may further send user features of multiple users to the slaves, for example, the master may broadcast the promotion item feature table X to all slaves, send user feature a of user a and user feature B of user B to the slave 1, send user feature C of user C and user feature D of user D to the slave 2, and send user feature E of user E and user feature F of user F to the slave 3.

For the slave end 2, the slave end 2 can combine a plurality of user characteristics and the promotion item characteristic table to generate a user promotion item characteristic table or a matrix; for example, a three-dimensional matrix is generated according to the promotion property feature table, the user feature c and the user feature d, wherein the three-dimensional matrix can be divided into three layers, namely a promotion property feature layer, a user feature c layer and a user feature d, and finally the three-dimensional matrix containing the promotion property features of the users is obtained.

For example, in some embodiments, in order to avoid data expansion problems and increase prediction speed caused when a Spark platform adopts an application program interface (Spark-API) to predict through an XGBoost algorithm, a JNI-API inside the Spark platform may be directly called to generate a data matrix (D-matrix) as a fusion feature according to a player feature of a player and a plurality of prop-discount features.

In some embodiments, the probability of a player's purchase of a game item at a discount may be predicted at the slave pair according to a data matrix; in some embodiments, to further improve the prediction speed, a plurality of data matrixes can be spliced into a large data matrix at the slave end, and the purchase probability of a plurality of players for game items under a plurality of discounts can be predicted for the large data matrix.

It should be noted that, in some embodiments, in order to increase the training speed, the master-slave mode of the Spark platform may also be used to predict the purchase probability in the model training phase.

Then, the subordinate terminal can adopt a prediction model to predict the purchase probability of the user for purchasing different promotion items according to the characteristics (namely, the fusion characteristics) of the promotion items of the user.

In some embodiments, the purchase probability of the user for purchasing the virtual item under the preset promotion scheme can be predicted in the slave terminal according to the extracted fusion features; in some embodiments, the purchase probability of the user for purchasing the virtual item under the preset promotion scheme can be predicted in the master control end according to the fusion characteristics returned by the slave end; in some embodiments, a master-slave architecture is not adopted, the fusion features are directly extracted locally, the purchase probability of the user for purchasing the virtual item under the preset promotion scheme is predicted according to the fusion features, and the like.

In some embodiments, in the model prediction stage, the property characteristics and the promotion scheme characteristics may be combined in pairs to obtain the property-discount characteristics, that is, the game properties with different discounts are regarded as different game properties, so that the property characteristics and the promotion scheme characteristics may be combined in pairs by the master end to obtain the property-discount characteristics, and the property-discount characteristics are distributed to the slave end in a small data broadcast manner, so as to avoid a (shuffle) operation, thereby increasing the operation speed.

In this embodiment, the target item may be determined in the virtual items based on the purchase probability, and the target promotion scheme may be determined in the preset promotion scheme.

The specific method may refer to step 104, which is not described herein.

Because the number of users is huge, the game props and promotion schemes are few and limited, by broadcasting the characteristics of the MxN promotion props to all the subordinate terminals and distributing the user characteristics of a large number of users to different subordinate terminals, a plurality of subordinate terminals can simultaneously carry out characteristic combination on the user characteristics and the promotion prop characteristics to obtain the fusion characteristics of the users, thereby predicting the purchase probability of the users for purchasing the promotion props according to the fusion characteristics, then determining the target props and the target promotion schemes based on the purchase probability, and enabling the master control terminal to efficiently and quickly obtain the target props and the target promotion schemes corresponding to each user, therefore, the scheme can save the prediction time and improve the recommendation efficiency.

204. And determining the final promotion scheme of each player according to the player distribution information corresponding to the target promotion scheme.

The distribution of different promotion schemes among players is often adjusted according to market demands in electronic games, for example, in some embodiments, a 5-fold item is recommended to 20% of players in the game, a 6-fold item is recommended to 30% of players in the game, and a 7-fold item is recommended to 50% of players in the game; the player distribution information may include information that different promotional programs are distributed among the players.

According to the player distribution information corresponding to the target promotion scheme, the final promotion scheme of each player can be determined.

For example, when it is required to recommend an item of 5 folds to 20% of players in the game, an item of 6 folds to 30% of players in the game, and an item of 7 folds to 50% of players in the game, all players may be ranked according to the total expected value purchased by each player, a player with the smallest total expected value is pushed out for 7 folds, a player with the largest total expected value is pushed out for 5 folds, the remaining 30% are pushed out for 6 folds, and so on.

Further, it is also possible to draw 7 out for the players whose total expected value is the largest 50%, 5 out for the players whose total expected value is the smallest 20%, and 6 out for the remaining 30%, and so on.

In this embodiment, when the slave performs the step "determine the target item in the virtual item based on the purchase probability, and determine the target promotion scheme in the preset promotion scheme", the final promotion scheme of each player is determined according to the player distribution information corresponding to the target promotion scheme.

For example, the subordinate terminal may perform the following steps to determine the final promotion scheme of each player according to the player distribution information corresponding to the target promotion scheme:

calculating the purchase expectation of the user for purchasing the promotion props according to the original price information, the promotion information and the purchase probability;

determining a promotion item adopting a preset promotion scheme in the M virtual items based on the purchase probability of the user for purchasing the promotion item;

determining the total purchase expectation of the target props corresponding to the preset promotion scheme according to the purchase expectation of all promotion props adopting the preset promotion scheme;

determining a target promotion scheme in the N preset promotion schemes based on the total purchase expectation corresponding to each preset promotion scheme;

and determining a target prop from the M virtual props.

For example, for a 7-fold prop a, a 7-fold prop B, and a 7-fold prop C adopting a 7-fold promotion scheme, the slave terminal 1 calculates a purchase expectation a, a purchase expectation B, and a purchase expectation C corresponding to the user, respectively, to obtain a total purchase expectation a + B + C; the slave terminal 2 calculates the corresponding purchase expectation D, purchase expectation E and purchase expectation F of the user for the prop D, the prop E and the prop F with 8 folds of the promotion scheme with 8 folds respectively to obtain a total purchase expectation D + E + F, and the like.

The slave end can send the total purchase expectation obtained by calculation to the master end, the master end sorts the users according to the total purchase expectation of each user to obtain sorted users, then the users which accord with the user distribution information are taken as target users from the sorted users, then the target promotion scheme is determined in N preset promotion schemes based on the total purchase expectation corresponding to each preset promotion scheme, and finally the target prop is determined in M virtual props.

In addition, the slave end can also send the total purchase expectation obtained by calculation to the master end, the master end collects and sends the total purchase expectation to a slave end specially used for sequencing, the sequencing slave end sequences the users according to the total purchase expectation of each user to obtain sequenced users, then the users meeting the user distribution information are taken from the sequenced users as target users, then a target promotion scheme is determined in N preset promotion schemes based on the total purchase expectation corresponding to each preset promotion scheme, finally a target prop is determined in M virtual props, and then the sequencing slave end sends the target promotion scheme and the target prop to the master end, and the like.

Specifically, based on the total purchase expectation corresponding to each preset promotion scheme, the method for determining the target promotion scheme among the N preset promotion schemes and determining the target prop among the M virtual props may refer to step 104, which is not described herein again.

205. And sending the corresponding target prop and the final promotion scheme to the client of each player so that the client of each player can display a shopping mall recommendation page according to the target prop and the final promotion scheme.

Finally, the target item and the final promotion scheme corresponding to the target item can be sent to the client of each player, so that the client of each player can display a shopping mall recommendation page according to the target item and the final promotion scheme.

For example, referring to fig. 2B1, the target property of player a is property a, property B, property C, property D, and the final promotion scheme is a 5-fold scheme, so the client, after receiving the target property and the final promotion scheme, can display the mystery store page of the property store, on which the final promotion scheme "my lucky discount is 5-fold", and the images, original price, 5-fold price, etc. of property a, property B, property C, property D can be displayed.

For example, referring to fig. 2B2, the target property of player B is property E, property F, property G, property H, and the final promotion scheme is a 7-fold scheme, so the client, after receiving the target property and the final promotion scheme, can display the mystery store page of the property store, on which the final promotion scheme "my lucky discount is 7-fold", and the images, original price, 7-fold price, etc. of property E, property F, property G, property H can be displayed.

By recommending the game property through the scheme, the conversion rate of the number of paid players and the Average income of each paid player (ARPPU) can be effectively improved.

Referring to fig. 2c, fig. 2c is a comparison graph generated by recommending a target property to a player by using a target promotion scheme and recommending a property to a player by using an original scheme in the scheme, in which the conversion rate of the number of paid players is increased by 85% and the ARPPU is increased by 2% compared with the original scheme.

Referring to fig. 2d, fig. 2c is a comparison graph generated by recommending target properties to players by adopting the final promotion scheme and recommending the properties to players by adopting the original scheme, compared with the original scheme, the conversion rate of the number of paid players is improved by 40%, and the conversion rate of ARPPU is improved by 10%.

As can be seen from FIGS. 2c and 2d, the effect of recommending the target item to the player is stronger than that of the original scheme, regardless of the target promotion scheme or the final promotion scheme.

According to the scheme, the training sample set and the prediction sample set can be obtained; training the preset model by adopting a training sample set until the preset model converges to obtain a prediction model; predicting the prediction sample set by adopting a prediction model, and determining the purchase probability of each player purchasing different game properties in the prediction sample set under a promotion scheme; determining a target promotion scheme for each player based on the purchase probability; determining a final promotion scheme of each player according to the player distribution information corresponding to the target promotion scheme; and sending the corresponding target prop and the final promotion scheme to the client of each player so that the client of each player can display a shopping mall recommendation page according to the target prop and the final promotion scheme.

In a game item recommendation scene, factors influencing a player to purchase an item may include preference and demand of the player on the item, payment capability of the player, discount strength of the item, and the like, and the prediction model provided by this embodiment may learn preference of a player on a certain game item under a certain promotion scheme.

Therefore, accuracy of the item recommendation method is improved.

In order to better implement the method, an embodiment of the present invention further provides a property recommendation device, where the property recommendation device may be specifically integrated in an electronic device, and the electronic device may be a terminal, a server, or other devices. The terminal can be a mobile phone, a tablet computer, an intelligent Bluetooth device, a notebook computer, a personal computer and other devices; the server may be a single server or a server cluster composed of a plurality of servers.

In some embodiments, the item recommendation device may be specifically integrated in a server cluster, where the server cluster may include a master server and a slave server.

The property recommending device integrated in the slave end server can receive the user characteristics and the MxN promotional property characteristics from the master control end; combining the characteristics of the user and the characteristics of the MxN promotion props to obtain MxN fusion characteristics of the user, wherein the fusion characteristics are characteristics of the user purchasing the promotion props; predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics; determining a target item in the M virtual items based on the purchase probability of the user for purchasing the promotion item, and determining a target promotion scheme in the N preset promotion schemes; and returning the target prop and the target promotion scheme of the user to the master control end.

Wherein, in some embodiments, the property promotion information can include the property information of virtual property and the promotion information of presetting the promotion scheme, and the property recommendation device integrated in the subordinate server can be used for extracting the characteristics of user information, property information and promotion information when obtaining the fusion characteristics:

receiving user characteristics and a promotion prop characteristic table from a master control end, wherein the promotion prop characteristic table comprises M multiplied by N promotion prop characteristics;

generating a user promotion prop characteristic table based on the user characteristics and the promotion prop characteristic table, wherein the user promotion prop characteristic table comprises M multiplied by N fusion characteristics;

at this time, in some embodiments, the item recommendation device integrated in the slave server may be further configured to, after predicting the purchase probability of the user purchasing the virtual item under the preset promotion scheme according to the fusion feature:

determining a target item and a target promotion scheme based on the purchase probability;

and returning the target prop and the target promotion scheme to the master control end.

For example, in this embodiment, the method of the embodiment of the present invention will be described in detail by taking an example that the item recommendation device is specifically integrated in a server cluster.

For example, as shown in fig. 3, the item recommendation apparatus may include an obtaining unit 301, a feature unit 302, a combining unit 303, a fusing unit 304, a predicting unit 305, a determining unit 306, and a recommending unit 307, as follows:

the acquisition unit 301:

the obtaining unit 301 may be configured to obtain user information of a user, M pieces of virtual item information, and N pieces of preset promotion scheme information, N, M all being positive integers greater than or equal to 1.

Feature cell 302:

the feature unit 302 may be configured to perform feature extraction on the user information, the M pieces of virtual item information, and the N pieces of preset promotion scheme information, to obtain a user feature, M pieces of item features, and N pieces of promotion features.

(iii) combining unit 303:

the combination unit 303 may be configured to perform feature combination on the M prop features and the N promotion features to obtain M × N promotion prop features, where the promotion prop features are features of promotion props, and the promotion props are virtual props for promotion by using a preset promotion scheme.

(iv) fusion unit 304:

the fusion unit 304 may be configured to perform feature combination on the user feature and the mxn promotional item features to obtain mxn fusion features of the user, where the fusion features are features of the user purchasing the promotional item.

(V) prediction Unit 305:

prediction unit 305 may be configured to predict a purchase probability of the user purchasing the promotional item according to the fused features.

In some embodiments, prediction unit 305 may include an acquisition subunit, a training subunit, and a prediction subunit, as follows:

(1) an acquisition subunit:

the obtaining subunit may be configured to obtain a training sample set, where the training sample set includes a plurality of training samples labeled with real purchase labels of training users, and the training samples include training fusion features, where the training fusion features are features of training users purchasing training promotion props.

In some embodiments, the acquisition subunit may be to:

In some embodiments, the acquisition subunit further comprises an unpurchased submodule and a filter submodule, wherein:

the non-purchase submodule can be used for determining training samples marked as non-purchase types in the training sample set as non-purchase training samples;

the screening submodule can be used for screening the training samples which are not purchased by the training users in the training sample set to obtain the screened training sample set.

In some embodiments, the screening submodule may be configured to:

performing user clustering processing on training users based on the training user information to obtain a plurality of training user clusters;

determining a training sample corresponding to each training user in a training sample set in a training user cluster;

and screening the training sample corresponding to each training user in the training sample set in the training user cluster to obtain the screened training sample set corresponding to the training user cluster.

(2) A training subunit:

the training subunit can be used for training the preset model by adopting the training sample set until the preset model converges to obtain the prediction model.

In some embodiments, the preset model includes a plurality of tree nodes, the training sample set includes a plurality of training samples, the training samples include real purchase labels, and the training submodule may be configured to:

In some embodiments, the training subunit may be configured to train the preset model using the filtered training sample set until the preset model converges to obtain the prediction model.

(3) A predictor unit:

the prediction subunit may be configured to predict, by using the prediction model, a purchase probability of the user purchasing the virtual item under the preset promotion scheme according to the fusion feature.

(sixth) determination unit 306:

determining unit 306 may be configured to determine a target item among M virtual items based on a purchase probability of a user purchasing a promotional item, and determine a target promotional scheme among N preset promotional schemes.

In some embodiments, the item information may include original price information of the virtual item, and the determining unit 306 may include an expectation subunit, an item determination subunit, a total expectation subunit, a target scheme determination subunit, and a target item determination subunit, as follows:

(1) the desired subunit:

the expectation subunit may be configured to calculate a purchase expectation of the user to purchase the promotional item according to the original price information, the promotion information, and the purchase probability.

In some embodiments, the desired subunit may be used to:

(2) A prop determination subunit:

the item determining subunit may be configured to determine, based on a purchase probability of the user purchasing the promotional item, a promotional item that adopts a preset promotional scheme among the M virtual items.

In some embodiments, the prop-determining subunit may be operable to:

acquiring historical purchase quantity of a user;

and determining the number of previous historical purchased promotion props adopting a preset promotion scheme from the M virtual props according to the purchase probability of the users for purchasing the promotion props.

(3) Total desired subunits:

the total expectation subunit may be configured to determine, according to the purchase expectation of all the promotional items using the preset promotional scheme, a purchase total expectation corresponding to the preset promotional scheme.

In some embodiments, the total desired subunit may be used to:

and accumulating the purchase expectation of the previous historical purchased number of the promotion items to obtain the purchase expectation of all the promotion items of the preset promotion scheme and determine the purchase total expectation corresponding to the preset promotion scheme.

In some embodiments, the overall desired subunit is further for:

and taking the users which accord with the user distribution information from the sorted users as target users.

(4) The target scheme determination subunit:

the target proposal determination subunit may be configured to determine the target promotion scheme among the N preset promotion schemes based on the total purchase expectation corresponding to each preset promotion scheme.

(5) Target prop determination subunit:

the target item determination subunit may be configured to determine the target item among the M virtual items.

(seventh) recommendation unit 307:

recommendation unit 307 may be configured to recommend the target item to the user using the target promotional program.

In some embodiments, the recommendation unit 307 may include:

and recommending the target prop to the target user by adopting a target promotion scheme.

As can be seen from the above, in the item recommendation device of this embodiment, the obtaining unit obtains the user information of the user, the M pieces of virtual item information, and the N pieces of preset promotion scheme information, N, M are positive integers greater than or equal to 1; the characteristic unit extracts the characteristics of the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics; the combination unit performs characteristic combination on the M prop characteristics and the N promotion characteristics to obtain M multiplied by N promotion prop characteristics, wherein the promotion prop characteristics are characteristics of promotion props, and the promotion props are virtual props which are promoted by adopting a preset promotion scheme; the fusion unit performs characteristic combination on the user characteristics and the MXN promotion item characteristics to obtain MXN fusion characteristics of the user, and the fusion characteristics are characteristics of the user for purchasing the promotion items; predicting the purchase probability of the user for purchasing the promotion item according to the fusion characteristics by a prediction unit; determining, by a determining unit, a target item among the M virtual items based on a purchase probability of a user purchasing the promotional item, and a target promotional scheme among the N preset promotional schemes; and recommending the target prop to the user by the recommending unit by adopting a target promotion scheme.

Therefore, the accuracy of the item recommendation method can be improved.

The embodiment of the invention also provides the electronic equipment which can be equipment such as a terminal, a server and the like. The terminal can be a mobile phone, a tablet computer, an intelligent Bluetooth device, a notebook computer, a personal computer and the like; the server may be a single server, a server cluster composed of a plurality of servers, or the like.

In this embodiment, a detailed description will be given by taking an example that the electronic device of this embodiment is a server, for example, as shown in fig. 4, it shows a schematic structural diagram of a server according to an embodiment of the present invention, specifically:

the server may include components such as a processor 401 of one or more processing cores, memory 402 of one or more computer-readable storage media, a power supply 403, an input module 404, and a communication module 405. Those skilled in the art will appreciate that the server architecture shown in FIG. 4 is not meant to be limiting, and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components. Wherein:

the processor 401 is a control center of the server, connects various parts of the entire server using various interfaces and lines, and performs various functions of the server and processes data by running or executing software programs and/or modules stored in the memory 402 and calling data stored in the memory 402, thereby performing overall monitoring of the server. In some embodiments, processor 401 may include one or more processing cores; in some embodiments, processor 401 may integrate an application processor, which primarily handles operating systems, user interfaces, applications, etc., and a modem processor, which primarily handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 401.

The memory 402 may be used to store software programs and modules, and the processor 401 executes various functional applications and data processing by operating the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to the use of the server, and the like. Further, the memory 402 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 access to the memory 402.

The server also includes a power supply 403 for supplying power to the various components, and in some embodiments, the power supply 403 may be logically connected to the processor 401 via a power management system, so that the functions of managing charging, discharging, and power consumption are implemented via the power management system. The power supply 403 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.

The server may also include an input module 404, the input module 404 operable to receive entered numeric or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

The server may also include a communication module 405, and in some embodiments the communication module 405 may include a wireless module, through which the server may wirelessly transmit over short distances to provide wireless broadband internet access to the user. For example, the communication module 405 may be used to assist a user in sending and receiving e-mails, browsing web pages, accessing streaming media, and the like.

Although not shown, the server may further include a display unit and the like, which will not be described in detail herein. Specifically, in this embodiment, the processor 401 in the server loads the executable file corresponding to the process of one or more application programs into the memory 402 according to the following instructions, and the processor 401 runs the application program stored in the memory 402, thereby implementing various functions as follows:

acquiring user information, M pieces of virtual prop information and N pieces of preset promotion scheme information of a user, wherein N, M are positive integers which are more than or equal to 1;

extracting characteristics of the user information, the M pieces of virtual prop information and the N pieces of preset promotion scheme information to obtain user characteristics, M pieces of prop characteristics and N pieces of promotion characteristics;

and recommending the target prop to the user by adopting a target promotion scheme.

The above operations can be implemented in the foregoing embodiments, and are not described in detail herein.

Therefore, the accuracy of the property recommendation method can be improved.

It will be understood by those skilled in the art that all or part of the steps of the methods of the above embodiments may be performed by instructions or by associated hardware controlled by the instructions, which may be stored in a computer readable storage medium and loaded and executed by a processor.

To this end, the embodiment of the present invention provides a computer-readable storage medium, in which a plurality of instructions are stored, where the instructions can be loaded by a processor to execute the steps in any item recommendation method provided by the embodiment of the present invention. For example, the instructions may perform the steps of:

Wherein the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

Since the instructions stored in the storage medium can execute the steps in any item recommendation method provided in the embodiment of the present invention, beneficial effects that can be achieved by any item recommendation method provided in the embodiment of the present invention can be achieved, which are detailed in the foregoing embodiments and will not be described herein again.

The method, the apparatus, the server and the computer-readable storage medium for recommending the road furniture provided by the embodiments of the present invention are described above, and the principle and the implementation of the present invention are explained in this document by applying specific embodiments, and the description of the above embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for those skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims

1. A method for recommendation of a utensil, comprising:

2. The item recommendation method of claim 1, wherein said item information comprises original price information of said virtual items, said determining target items among said M virtual items based on purchase probabilities of said user purchasing said promotional items, and determining target promotional schemes among said N preset promotional schemes, comprises:

calculating the purchase expectation of the user for purchasing the promotion prop according to the original price information, the promotion information and the purchase probability;

determining the promotion item adopting the preset promotion scheme in M virtual items based on the purchase probability of the user for purchasing the promotion item;

determining a total purchase expectation corresponding to the preset promotion scheme according to the purchase expectation of all promotion items adopting the preset promotion scheme;

and determining a target prop in the M virtual props.

3. The item recommendation method according to claim 2, wherein the calculating the purchase expectation of the user for the virtual item under the preset promotion scheme according to the original price information, the promotion information and the purchase probability comprises:

4. The item recommendation method of claim 2, wherein said determining, among M virtual items, a promotional item that employs the preset promotional scheme based on a probability of purchase of the promotional item by the user comprises:

acquiring historical purchase quantity of the user;

determining a total purchase expectation corresponding to the preset promotion scheme according to the purchase expectation of all promotion items adopting the preset promotion scheme, wherein the step comprises the following steps of:

5. The item recommendation method according to claim 2, wherein after determining the total purchase expectation corresponding to the preset promotional scheme according to the purchase expectation of all promotional items using the preset promotional scheme, the method further comprises:

6. The item recommendation method according to claim 1, wherein the predicting the purchase probability of the user to purchase the virtual item under the preset promotion scheme according to the fused features comprises:

acquiring a training sample set, wherein the training sample set comprises a plurality of training samples marked with real purchase marks of training users, the training samples comprise training fusion features, and the training fusion features are features of the training users for purchasing training promotion props;

training a preset model by adopting the training sample set until the preset model is converged to obtain a prediction model;

and predicting the purchase probability of the user for purchasing the virtual prop under the preset promotion scheme according to the fusion characteristics by adopting the prediction model.

7. The item recommendation method of claim 6, wherein said obtaining a training sample set comprises:

8. The item recommendation method of claim 6, wherein after obtaining the training sample set, further comprising:

determining the training samples marked as unpurchased types of the real purchases as unpurchased training samples in the training sample set;

screening the unpurchased training samples of the training users in the training sample set to obtain a screened training sample set;

the training of the preset model by adopting the training sample set until the preset model converges to obtain a prediction model comprises the following steps:

9. The property recommendation method of claim 8, wherein said filtering the unpurchased training samples of the training users in the training sample set to obtain a filtered training sample set comprises:

10. The property recommendation method of claim 6, wherein the preset model comprises a plurality of tree nodes, the training sample set comprises a plurality of the training samples, the training samples comprise real purchase labels, and the training of the preset model using the training sample set until the preset model converges to obtain a prediction model comprises:

11. A ballast recommendation device, comprising:

12. A recommendation system is characterized by comprising a master control end and a slave end, wherein,

the main control end is used for executing the following steps:

the slave is used for executing the following steps:

acquiring user characteristics and MxN promotional prop characteristics;

13. The recommendation system according to claim 12, wherein the slave, when obtaining the user characteristics and the mxn promotional item characteristics, is configured to:

14. A server comprising a processor and a memory, the memory storing a plurality of instructions; the processor loads instructions from the memory to perform the steps of the item recommendation method according to any one of claims 1 to 11.

15. A computer-readable storage medium storing instructions adapted to be loaded by a processor to perform the steps of the item recommendation method according to any one of claims 1 to 11.