CN112785390B

CN112785390B - Recommendation processing method, device, terminal equipment and storage medium

Info

Publication number: CN112785390B
Application number: CN202110143085.9A
Authority: CN
Inventors: 张懿; 甘露; 吴伟佳; 洪庚伟; 李羽
Original assignee: Weimin Insurance Agency Co Ltd
Current assignee: Weimin Insurance Agency Co Ltd
Priority date: 2021-02-02
Filing date: 2021-02-02
Publication date: 2024-02-09
Anticipated expiration: 2041-02-02
Also published as: CN112785390A

Abstract

The embodiment of the application discloses a recommendation processing method, a recommendation processing device, terminal equipment and a storage medium, wherein the recommendation processing method comprises the following steps: acquiring fusion characteristic data of each element to be recommended in an element set to be recommended; sequencing and predicting each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended to obtain a predicted sequencing value of each element to be recommended; performing sequencing optimization processing based on historical feature data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended; and determining the recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value. By adopting the embodiment of the application, the recommendation accuracy can be improved, and the applicability is high.

Description

Recommendation processing method, device, terminal equipment and storage medium

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a recommendation processing method, a recommendation processing device, terminal equipment and a storage medium.

Background

With the continuous development of online shopping platforms, recommendation systems have become an important component of electronic commerce. The recommendation system can learn the hidden preference information in the historical behaviors of the user, so that the shopping behaviors of the user are further predicted, customers are helped to select satisfactory articles, and meanwhile, the income of the electronic commerce platform is promoted. At present, many recommendation systems recommend articles according to different crowds, however, even different individuals among the same crowd may have preference differences, so that the recommendation accuracy is lower.

Disclosure of Invention

The embodiment of the application provides a recommendation processing method, a recommendation processing device, terminal equipment and a storage medium, which can improve the recommendation accuracy and have high applicability.

The embodiment of the application provides a recommendation processing method, which comprises the following steps:

acquiring fusion characteristic data of each element to be recommended in the element set to be recommended, wherein the fusion characteristic data comprises user characteristic data of a target recommended user;

sequencing and predicting each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended to obtain a predicted sequencing value of each element to be recommended;

Performing sequencing optimization processing based on historical feature data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended, and the historical feature data comprises user feature data of a sample user;

and determining the recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value.

The embodiment of the application provides a recommendation processing device, which comprises:

the fusion characteristic data determining module is used for acquiring fusion characteristic data of each element to be recommended in the element set to be recommended, wherein the fusion characteristic data comprises user characteristic data of a target recommended user;

the prediction ranking value determining module is used for performing ranking prediction on each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended to obtain a prediction ranking value of each element to be recommended;

the reordering factor acquisition module is used for carrying out ordering optimization processing based on the historical characteristic data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended, and the historical characteristic data comprises user characteristic data of a sample user;

The recommendation ranking value determining module is used for determining the recommendation ranking value of each element to be recommended according to the prediction ranking value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ranking order of each element to be recommended in the element set to be recommended according to the recommendation ranking value.

The embodiment of the application provides terminal equipment, which comprises a processor and a memory, wherein the processor and the memory are connected with each other. The memory is for storing a computer program supporting the terminal device to perform the method provided in the first aspect and/or any of the possible implementation manners of the first aspect, the computer program comprising program instructions, the processor being configured to invoke the program instructions to perform the method provided above.

The present embodiments provide a computer readable storage medium storing a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the method provided above.

In the embodiment of the application, fusion characteristic data of each element to be recommended in an element set to be recommended is obtained; sequencing and predicting each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended to obtain a predicted sequencing value of each element to be recommended; performing sequencing optimization processing based on historical feature data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended; and determining the recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value. By adopting the embodiment of the application, the recommendation accuracy can be improved, and the applicability is high.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed 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 application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a system architecture according to an embodiment of the present application;

FIG. 2 is a schematic flow chart of a recommendation processing method according to an embodiment of the present application;

FIG. 3 is a schematic flow chart of an algorithm provided in an embodiment of the present application;

FIG. 4 is a schematic diagram of offline verification results of a reordering algorithm provided in an embodiment of the present application;

FIG. 5 is a schematic diagram of the verification results of the AB test provided in the examples of the present application;

FIG. 6 is a schematic structural diagram of a recommendation processing apparatus according to an embodiment of the present disclosure;

FIG. 7 is another schematic structural diagram of a recommendation processing device according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of a terminal device provided in an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.

The recommendation processing method provided by the embodiment of the application can be widely applied to various online recommendation systems. For example, the online recommendation system may be a commodity recommendation system, a service recommendation system, a song recommendation system, a video recommendation system, and the like, which are specifically determined according to the actual application scenario, and are not limited herein. The service recommendation system may be a system for insurance service recommendation, and the like, and is not limited herein. For convenience of description, the present application refers to an object to be recommended as an element to be recommended, or an item to be recommended (item), or the like. The above-mentioned recommended articles are generalized articles, and are the general names of all recommended objects.

Referring to fig. 1, fig. 1 is a schematic diagram of a system architecture according to an embodiment of the present application. As shown in fig. 1, the server 100a, the server 100b, and the terminal device 101 may be connected to each other via a network. The terminal device shown in fig. 1 may include a mobile phone, a tablet computer, a notebook computer, a palm computer, a mobile internet device (mobile internet device, MID), a wearable device (e.g., a smart watch, a smart bracelet, etc.), and the like, which is not limited herein. It should be understood that the server 100a may perform data transmission with the server 100b, and the server 100a and the server 100b may also perform data transmission with the terminal device 101, respectively. The terminal device may be loaded with application programs having various functionalities, and for convenience of description, the application programs having various functionalities loaded on the terminal device may be exemplified by the first application. Alternatively, the first application may be a video application, a music playing application, a game application, a shopping application, or the like, or the first application may be an applet, a web page, or the like, which is not limited herein.

It is understood that the server 100a and the server 100b may be local servers of the first application, remote servers (e.g., cloud servers), or the like, and are not limited herein. The server 100a may be configured to store various types of data (e.g., user characteristic data of a registered user of the first application, element characteristic data of each element to be recommended, etc.) and information related to the first application. The server 100b may be configured to obtain various feature data required for the recommendation task from the server 100a, and perform processing according to the feature data to generate a recommendation list, so as to send the recommendation list to the terminal device for display. That is, when the first application in the terminal device transmits a data acquisition request to the server 100b, the server 100b may return a data instance (e.g., a recommendation list as in fig. 1) to the first application in the terminal device based on the received data acquisition request to update the interface presentation instance of the first application in the terminal device.

The method in the embodiment of the application comprises the steps of obtaining fusion characteristic data of each element to be recommended in an element set to be recommended; sequencing and predicting each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended to obtain a predicted sequencing value of each element to be recommended; performing sequencing optimization processing based on historical feature data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended; and determining the recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value. By adopting the embodiment of the application, the recommendation accuracy can be improved.

The method and the related apparatus according to the embodiments of the present application will be described in detail below with reference to fig. 2 to 7, respectively.

Referring to fig. 2, fig. 2 is a flowchart illustrating a recommended processing method according to an embodiment of the present application. The method provided by the embodiment of the application may include the following steps S201 to S204:

s201, acquiring fusion characteristic data of each element to be recommended in the element set to be recommended.

In some possible embodiments, by acquiring the user feature data of the target recommended user and the element feature data of each element to be recommended in the set of elements to be recommended, fusion feature data corresponding to each element to be recommended may be generated according to the user feature data of the target recommended user and the element feature data of each element to be recommended. The target recommending user is a directivity object of the current element recommendation/article recommendation. For example, if the current item recommendation is directed to the user1, the user1 is the target recommended user, and if the current item recommendation is directed to the user2, the user2 is the target recommended user. For convenience of description, the following embodiments of the present application will take the target recommended user as an example. It will be appreciated that the target recommendation user may be a registered user in the recommendation system, wherein the registered user may be a newly registered user, such as a user who has not purchased or has a record of the order, or may be an old registered user, such as a user who has purchased or has a record of the order, and the like, which is not limited herein. Wherein the user characteristic data of the target recommended user includes at least one of a user gender, a user age, a user browsing history, a user purchase history, and the like. It can be understood that at least two elements to be recommended included in the element set to be recommended in the present application are elements under the same recommended task scene. For example, a recommended task scenario may be a special offer scenario, or a payment completion scenario, or an insurance list scenario, or a child product scenario, etc., without limitation. The element to be recommended may also be referred to as an item to be recommended, and the element characteristic data of the element to be recommended may be referred to as item characteristic data of the item to be recommended. It should be appreciated that the item to be recommended for any item to be recommended includes at least one of an item identification, an item type, an item price, an item manufacturer, and the like.

It can be understood that, in the present application, the feature data of each dimension included in the user feature data and the item feature data may be converted into feature data in the libvm format and then processed. That is, the present application may measure user feature data or element feature data of each dimension in a libvm manner. For example, regarding the gender in the user feature data, the gender "man" may be converted into a form of (gender_man: 1), and for the browsing history, assuming that the browsed product is item a, item b, and the unbrown product is item c, the browsing history feature may be configured as (browsing history_item a:1, browsing history_item b:1, browsing history_item c: 0), without limitation.

In some possible embodiments, the fusion characteristic data of one element to be recommended in the element set to be recommended and the user characteristic data of the target recommendation user are spliced to obtain the fusion characteristic data of the one element to be recommended, and further, fusion characteristic data corresponding to each element to be recommended in the element set to be recommended can be generated. The splicing mode of the user characteristic data and the element characteristic data in the fusion characteristic data can be determined according to the actual application scene, and the method is not limited. For example, assuming that the target recommended user is user1, the user feature data corresponding to user1 is feature1, and the element feature data of any element item a to be recommended is feature2, the fusion feature data ra= [ feature1, feature2] corresponding to the element to be recommended, or the fusion feature data ra= [ feature2, feature1] corresponding to the element to be recommended is not limited herein.

S202, sequencing and predicting each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended, and obtaining a predicted sequencing value of each element to be recommended.

In some possible embodiments, the predicted ranking value of each element to be recommended may be obtained by performing ranking prediction on each element to be recommended in the set of elements to be recommended according to the fusion feature data of each element to be recommended. Specifically, the embodiment of the application can input the fusion characteristic data of each element to be recommended into a ranking value prediction model, and perform ranking prediction on each element to be recommended in the element set to be recommended through the ranking value prediction model to obtain a predicted ranking value of each element to be recommended. The ranking value prediction model referred to in the present application may be a Conversion Rate (CVR) prediction model, or a Click-Through-Rate (CTR) prediction model, etc., which is not limited herein. When the ranking value prediction model is a CVR prediction model, the predicted ranking value is a predicted conversion rate, and correspondingly, when the ranking value prediction model is a CTR prediction model, the predicted ranking value is a predicted click rate.

Generally, the conversion rate of an item is the ratio of the conversion rate of the item to the click rate, and the click rate of the item is the ratio of the click rate of the item to the display rate. For ease of understanding, the concept of click-through rate and conversion rate will be described with an off-line brick and mortar store as an example. For example, suppose that in a promotion, a merchant printed 10000 leaflets, all sent out, all the 10000 people seen, and the actual number of people in store was 1000. Then 10000 people are displayed and 1000 people enter the store are clicked, so the click rate is 1000/1000×100% =10%. Further, assuming that after the 1000 persons enter the store, 50 persons purchase the order, the conversion rate is 50/1000×100+=5%.

It can be understood that if the training sample for training the initial prediction model is user feature data of a sample user, element feature data of an element to be recommended, and a historical actual conversion rate corresponding to the element to be recommended, the ranking value prediction model is a conversion rate prediction model. Correspondingly, if the training sample for training the initial prediction model is user characteristic data of a sample user, element characteristic data of an element to be recommended and a historical actual click rate corresponding to the element to be recommended, the ranking value prediction model is a click rate prediction model. The sample user is an old user in the recommendation system, and may or may not include the target recommendation user, and is specifically determined according to an actual application scenario, which is not limited herein. For example, if the target recommended user is a newly registered user, the sample user does not include the target recommended user, and if the target recommended user is an old user, the sample user may include the target recommended user.

For ease of understanding, the training process of the conversion rate prediction model will be described herein by taking the ranking value prediction model as an example of the conversion rate prediction model. Before model training, a training sample set is obtained, whereinMultiple training samples may be included. The training sample comprises user characteristic data of a sample user, element characteristic data of an element to be recommended and historical actual conversion rate corresponding to the element to be recommended. When model training is performed, the user characteristic data and the element characteristic data in each training sample are input into the initial prediction model, so that the prediction conversion rate of each training sample output by the initial prediction model can be obtained. Further, a difference (i.e., a loss function) between the predicted conversion rate of each training sample outputted by the initial prediction model and the historical actual conversion rate included in each training sample is calculated, so that model parameters of the initial prediction model are continuously adjusted according to the difference between the predicted conversion rate of each training sample and the historical actual conversion rate, and the adjusted initial prediction model is determined as a conversion rate prediction model until the adjusted initial prediction model converges. It is to be appreciated that the initial prediction model may be a logistic regression (Logistic Regression, LR) model, an XGBOOST model, a neural network model, or the like, which is specifically determined according to an actual application scenario, and is not limited herein. Specifically, the embodiment of the application may use cross entropy as the loss function during model training, that is, the loss function in the embodiment of the application satisfies: Wherein y is _i Representing the historical actual conversion rate, p, corresponding to the elements to be recommended included in any training sample _i And representing the predicted conversion rate corresponding to the element to be recommended, which is output by the conversion rate prediction model. It can be understood that if the sum of the calculated cross entropy is smaller than or equal to the preset cross entropy, the conversion rate prediction model obtained through training is proved to be in accordance with the construction requirement, otherwise, the conversion rate prediction model obtained through training is proved to be out of accordance with the construction requirement, and the training of the conversion rate prediction model is continued until the conversion rate prediction model meets the construction requirement.

And S203, sorting optimization processing is carried out based on the historical characteristic data of each element to be recommended in the element set to be recommended, and a reordering factor sequence is obtained.

In some possible embodiments, the ranking optimization process is performed based on the historical feature data of each element to be recommended in the element set to be recommended, so as to obtain a ranking factor sequence. Specifically, the historical characteristic data of each element to be recommended in the element set to be recommended is input into the ranking value prediction model, so that a historical ranking value set corresponding to each element to be recommended can be obtained, wherein the historical ranking value set corresponding to any element to be recommended comprises: and N historical ranking values of any element to be recommended to N sample users are obtained, wherein N is an integer greater than 0. Therefore, the reordering optimization processing is performed based on the historical ordering value sets corresponding to the elements to be recommended, and a reordering factor sequence can be obtained. It will be appreciated that the above-mentioned one history feature data includes element feature data of one element to be recommended and user feature data of one sample user. That is, the historical characteristic data of an element to be recommended is obtained by splicing the element characteristic data of the element to be recommended with the user characteristic data of a sample user. For example, assuming that a certain sample user is user2, the user feature data corresponding to user2 is feature0, and the element feature data of the element item a to be recommended is feature2, the historical feature data Ra '= [ feature0, feature2] corresponding to the element to be recommended, or the historical feature data Ra' = [ feature2, feature0] corresponding to the element to be recommended is not limited herein. The data format of the fusion characteristic data and the data format of the history characteristic data are kept consistent. That is, if the splicing order of the user feature data and the element feature data is specified as [ user feature data, element feature data ], the data format of the history feature data is [ user feature data of sample user, element feature data of element to be recommended ], and the data format of the fusion feature data is [ user feature data of target recommended user, element feature data of element to be recommended ]. If the splicing sequence of the user characteristic data and the element characteristic data is [ element characteristic data, user characteristic data ], the data format of the history characteristic data is [ element characteristic data of the element to be recommended, user characteristic data of the sample user ], and the data format of the fusion characteristic data is [ element characteristic data of the element to be recommended, and the target recommends the user characteristic data of the user ].

It should be understood that, if there are N sample users in the present application, where N is an integer greater than 0 (i.e., there is at least one sample user in the present application), the N historical feature data corresponding to the one element to be recommended may be obtained by stitching the user feature data of each sample user in the N sample users with the element feature data of the one element to be recommended (for example, the element item a to be recommended). Therefore, the N historical ranking values corresponding to the element to be recommended can be obtained through the ranking value prediction model by inputting the N historical characteristic data corresponding to the element to be recommended into the ranking value prediction model obtained through the pre-training. Furthermore, for each element to be recommended in the element set to be recommended, N historical feature data corresponding to each element to be recommended, namely a historical ranking value set corresponding to each element to be recommended, can be obtained.

In some possible embodiments, the above-mentioned reordering optimization based on the historical ranking value set corresponding to each element to be recommended, and obtaining the reordering factor sequence may be understood as: and determining an equivalent ranking value H (I) of each element to be recommended according to the historical ranking value set of each element to be recommended, acquiring the total exposure number I (I) of each element to be recommended to N sample users, and acquiring the value P (I) of each element to be recommended. And determining the total conversion value Re v corresponding to the element set to be recommended according to the equivalent sorting value H (I) of each element to be recommended, the total exposure number I (I) of each element to be recommended to N sample users and the price P (I) of each element to be recommended. Further, a reorder factor sequence is determined based on the total transformation value Rev. Wherein, the total conversion value Rev satisfies:

Wherein, I (I) represents the total exposure number of each element I to be recommended to N sample users, H (I) represents the equivalent sorting value of each element I to be recommended, and P (I) represents the price of each element I to be recommended.

It will be appreciated that in determining the total exposure number I (I) of each element to be recommended to N sample users, thatAcquiring the exposure number I of each element to be recommended to each sample user u _u (i) And further determining the total exposure number I (I) of each element to be recommended to all sample users by the sum of the exposure numbers of each element to be recommended to each sample user. When the sorting sequence number of any element to be recommended corresponding to any sample user is smaller than or equal to M, the exposure number of any sample user to be recommended is equal to 1, and when the sorting sequence number of any element to be recommended corresponding to any sample user is larger than M, the exposure number of any sample user to be recommended is equal to 0, wherein M is determined according to the maximum exposure number. That is, the total exposure number I (I) of any one element I to be recommended to N sample users can be based on the independent exposure function I of each sample user _u (i) Summing, i.eWherein I is _u (i) The method meets the following conditions:

wherein rank (CR) _u (i) For a certain sample user u), the ranking result of the historical ranking value of a certain element i to be recommended in the historical ranking values corresponding to all the elements to be recommended is represented, and the range of the ranking result is a natural number of 1,2 and …. M represents the maximum exposure number of the current scene. Therefore, the above exposure function can be understood as that when the sorting result of the element I to be recommended is less than or equal to M, it is considered that it can be exposed, i.e. I _u (i) =1, otherwise, if the sorting result of the element I to be recommended is greater than M, I _u (i) =0. Wherein CR is _u (i) The values of (2) may be derived from the ranking value prediction model described above.

For example, assume that there are a total of 50 elements to be recommended in an online mall, where M is equal to 30, taking the sample user as user 1. For any element to be recommended of the 50 elements to be recommended (for example, taking any element to be recommended as item a for illustration), the feature1 of the user1 is compared with the elements to be recommendedThe element characteristic data feature2 of the element item a is spliced to obtain fusion characteristic data Ra= [ feature1, feature2 corresponding to the item a]. Therefore, 50 fusion characteristic data corresponding to 50 elements to be recommended can be obtained for 50 elements to be recommended. It can be understood that, by inputting the 50 fusion feature data corresponding to the 50 elements to be recommended into the prediction ranking value model, 50 historical ranking values corresponding to the 50 elements to be recommended can be obtained. Wherein, assuming that the above 50 historical ranking values are ranked in order from large to small, it is determined that the historical ranking value of item a is at the 20 th position (i.e. the ranking number of item a is 20), since 20 < M, the exposure number I of item a for user1 _user1 (item a) =1. Similarly, the exposure number of item a to each of the N sample users can be obtained. For example, taking n=3 as an example, assume that the 3 sample users include, in addition to the user 1, a user 2 and a user 3, where item a has an exposure number I to the sample user 2 _user2 (item a) =0, exposure number I of item a to sample user 3 _user2 (item a) =1, then for item a, the total exposure number of item a to the 3 sample users is equal to 2, i.e. I (item a) =i _user1 (item a)+I _user2 (item a)+I _user3 (item a) =2. Similarly, for each element to be recommended included in the element to be recommended set, the total exposure number of each element to be recommended to the N sample users can be obtained.

It can be appreciated that when the equivalent ranking value H (i) of each element to be recommended is determined according to the historical ranking value set of each element to be recommended, the associated information set of each element to be recommended may be obtained, where the associated information set of any element to be recommended includes: association information between any element to be recommended and each sample user of the N sample users. Acquiring an initial sorting factor sequence, and determining an equivalent sorting value H (i) of each element to be recommended according to the initial sorting factor of each element to be recommended in the initial sorting factor sequence, a historical sorting value set corresponding to each element to be recommended and an association information set of each element to be recommended.

Specifically, the association information between the any element to be recommended and each sample user of the N sample users includes the exposure number of the any element to be recommended to each sample user. Therefore, the equivalent ranking value H (i) of each element to be recommended satisfies:

wherein, gamma (i) represents the initial reordering factor corresponding to each element i to be recommended, CR _u (i) Representing the historical ranking value of each element to be recommended I for each sample user, I _u (i) Representing the number of exposures per element to be recommended per sample user, ρ (rank (γ (i)) CR _u (i) ) represents the conversion deviation coefficient of each element i to be recommended. Wherein the conversion deviation coefficient of each element to be recommended in each sample user comprises: and determining a coefficient from at least one candidate conversion deviation coefficient corresponding to each element to be recommended according to the sequence number of each element to be recommended in the element set to be recommended, wherein the sequence number of any element to be recommended is determined according to the product of the initial reordering factor corresponding to any element to be recommended and the historical ordering value.

That is, the conversion deviation coefficient ρ (rank (γ (i)) CR in the present application _u (i) And) is determined from at least one candidate conversion deviation factor corresponding to the element i to be recommended based on the product of the initial reordering factor and the historical ordering value. Wherein one exposure position corresponds to one candidate conversion deviation coefficient. The at least one candidate conversion deviation coefficient is determined according to at least one historical ranking value corresponding to the element i to be recommended at least one exposure position. Specifically, the conversion deviation coefficient corresponding to the element i to be recommended at any one of the at least one exposure position is a ratio of the historical ranking value corresponding to the element i to be recommended at any one exposure position to the historical ranking value corresponding to the first one of the at least one exposure position.

For example, assuming that the historical conversion rate of item a at the first exposure position of the user 1 is 0.7, the historical conversion rate at the second exposure position of the user 1 is 0.6, and the historical conversion rate at the third exposure position of the user 1 is 0.5, it is known that the conversion deviation coefficient of the item a at the first exposure position is 0.7/0.7=1, the conversion deviation coefficient at the second exposure position is 0.6/0.7=6/7, and the conversion deviation coefficient at the third exposure position is 0.5/0.7=5/7 according to the ratio of the historical ranking value of the element i to be recommended at any exposure position to the historical ranking value of the element i to be recommended at any exposure position.

As can be appreciated, by combiningSubstituted intoThe total conversion value Rev can be obtained to satisfy:

the formula of the total conversion value Rev can be known, and the formula only contains one unknown number gamma (i), so that the corresponding relation between gamma (i) and Rev can be directly learned. That is, a reorder factor sequence may be determined based on the total conversion value Rev: for example, it may be detected whether a sum of average revenue change values of respective elements to be recommended satisfies a stop condition. If yes, determining a new reordering factor sequence based on an updating rule of the reordering factor sequence and the initial reordering factor sequence, and performing ordering optimization processing based on the new reordering factor sequence. Wherein, the reordering factor updating rule is determined according to the dot division result of the first reordering factor sequence and the second reordering factor sequence, the learning rate alpha and the average income change vector delta R (t). In this application, the first reordering factor sequence refers to: and a reordering factor sequence obtained by the last iteration corresponding to the current initial reordering factor sequence. The second reordering factor sequence refers to: and the reordering factor sequence obtained by the previous iteration corresponds to the reordering factor sequence obtained by the previous iteration.

In other words, the present application can train to obtain the reordering factor sequence that maximizes the total transformation value Rev based on the reinforcement learning method of the strategy gradient. This is because the embodiments of the present application cannot acquire the optimal reorder factor sequence at a time, but may acquire the corresponding Rev value by continuously trying different reorder factor sequences. This process can be abstracted into a continuous decision process, i.e. updating the own policy based on environmental feedback.

First, the present application may define three variables of the learning process:

state S: the application uses the triplet { I (I), H (I), gamma (I) } to represent the state of the learning process, and each iteration can be abstracted into transition between states.

Action a: the present application defines a change in γ as an action, and the action vector at any one time t may be expressed as a (t) = { Δγ (0, t),..Δγ (N, t) }, where N represents the number of elements to be recommended, and Δγ (N, t) represents the change value of the reordering factor of item N by time t. According to the definition above, the reordering factor vector γ (t+1) at time t+1 can be expressed as: γ (t+1) =γ (t) +a (t).

Awards R: considering that the optimization objective of the present application is to optimize the total conversion value (i.e., the total conversion value), then naturally the present application defines the current total conversion value as the prize R. The conversion value of each user is considered to be equally important to the application at any moment, so that R of different users is not subjected to weight reduction at the same moment t during calculation. According to the definition above, the bonus function can be calculated by the following formula:

According to the definition, the application designs a learning algorithm based on strategy gradients. In the context of the present application, the present application uses the average revenue change as the present application update direction vector, namely:

ΔR(t)＝[ΔR(0,t),...,ΔR(N,t)]

where Δr (k, t) represents the average revenue improvement of the element to be recommended k over the time t, Δr (k, t) =r (k, t) - (k, t-1). Meanwhile, the application also uses the difference of gamma of two iterations to determine the update gradient. Based on the above derivation, the final updated formula of the present application is as follows:

wherein,the point division of the gamma vector between two iterations is represented, and alpha represents the learning rate. By moving γ (t) to the left of the formula, the present application can get an updated action for each iteration. Wherein, the stopping condition satisfies:

where ε is a predefined hyper-parameter.

That is, the present application can train to obtain a reordering factor sequence that maximizes the above-mentioned total transformation value Rev based on reinforcement learning method of strategy gradient. The reordering factor sequence comprises one reordering factor corresponding to each element to be recommended.

S204, determining a recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value.

In some possible embodiments, the recommendation ranking value of each element to be recommended may be determined according to the prediction ranking value corresponding to each element to be recommended and the reordering factor corresponding to each element to be recommended, and further, the recommendation ranking order of each element to be recommended in the element set to be recommended may be determined according to each recommendation ranking value. Specifically, in the embodiment of the present application, the product of the predicted ranking value corresponding to any element to be recommended and the reordering factor corresponding to the element to be recommended may be determined as the recommended ranking value of the element to be recommended, and so on, the recommended ranking value of each element to be recommended in the element set to be recommended may be obtained.

Referring to fig. 3, fig. 3 is a schematic flowchart of an algorithm provided in an embodiment of the present application. As shown in fig. 3, the predicted ranking value determined based on the ranking value prediction model may be obtained by inputting the fused feature data of each element to be recommended in the element set to be recommended into the ranking value prediction model, and the historical feature data of each element to be recommended in the element set to be recommended may be obtained by inputting the historical feature data of each element to be recommended into the ranking value prediction model, so that the reordering optimization process may be performed based on the historical ranking value set corresponding to each element to be recommended, thereby obtaining the reordering factor sequence. Finally, the recommended ranking value of each element to be recommended can be obtained by multiplying the predicted ranking value of each element to be recommended by a reordering factor.

For example, taking a recommendation value prediction model as an example of a conversion rate prediction model, assume that a target recommendation user is user 1, and the set of elements to be recommended includes an element item a to be recommended, an element item b to be recommended, and an element item c to be recommended. The element feature data of the element item a to be recommended, the element feature data of the element item b to be recommended and the element feature data of the element item c to be recommended are respectively fused with the user feature data of the user 1 to be recommended, so that fusion feature data 1 corresponding to the element item a to be recommended, fusion feature data 2 corresponding to the element item b to be recommended and fusion feature data 3 corresponding to the element item c to be recommended can be respectively obtained. And respectively inputting the fusion characteristic data 1, the fusion characteristic data 2 and the fusion characteristic data 3 into a conversion rate prediction model, so that the predicted conversion rate of the element item a to be recommended is 0.7, the predicted conversion rate of the element item b to be recommended is 0.65, and the predicted conversion rate of the element item c to be recommended is 0.32. Assuming that the reordering factor corresponding to item a is equal to 0.8, the reordering factor corresponding to item b is equal to 0.9, and the reordering factor corresponding to item a is equal to 0.5 in the obtained reordering factor sequence, the recommended ranking value of the element item a to be recommended is 0.7x0.8=0.56, the recommended ranking value of item b is 0.65x0.9=0.585, and the recommended ranking value of item c is 0.32 x0.5=0.16.

In some possible embodiments, the above-mentioned recommendation ranking value may be used to determine the ranking order of each element to be recommended in the set of elements to be recommended. For example, the to-be-recommended element sets may be ranked according to the order of the recommended ranking values of the to-be-recommended elements from large to small, so as to obtain the item ranking order of the to-be-recommended element sets. Furthermore, according to the number n of the display interfaces of the terminal equipment, the first n elements to be recommended in the article arrangement sequence are displayed in the display interfaces of the terminal equipment. The number m of the articles of the element set to be recommended, which is included in the recommendation system, is larger than or equal to the displayable number n of the display interface. For example, assuming that m is 100 and n is 10, the first 10 to-be-recommended elements of the 100 to-be-recommended elements in the order of the items may be sequentially displayed in a display interface of a recommendation system of the terminal device.

For example, assume that 3 elements to be recommended are included in the set of elements to be recommended, item a, item b, and item c, respectively. Wherein, the recommended ranking value of item a is 0.56, the recommended ranking value of item b is 0.585, and the recommended ranking value of item c is 0.16, and the ranking order of the elements to be recommended is item b, item a, item c can be determined according to the order of the recommended ranking values from large to small. Therefore, the display effect of the elements to be recommended in the display interface of the terminal equipment is item b, item a and item c in sequence from top to bottom.

In the embodiment of the application, fusion characteristic data corresponding to each element to be recommended is generated according to user characteristic data of a target recommended user and element characteristic data of each element to be recommended in an element set to be recommended. And determining the predicted ranking value corresponding to each element to be recommended according to the ranking value prediction model and the fusion characteristic data corresponding to each element to be recommended. Acquiring a reordering factor sequence, determining a recommended ordering value of each element to be recommended according to a predicted ordering value corresponding to each element to be recommended and a reordering factor corresponding to each element to be recommended in the reordering factor sequence, wherein the recommended ordering value is used for determining the ordering sequence of the element set to be recommended. By adopting the embodiment of the application, the recommendation accuracy can be improved, and the applicability is high.

As can be appreciated, the inventor of the application also finds that, for different target recommendation users, after ranking each element to be recommended included in the element to be recommended set based on the recommendation processing method provided by the application, the recommendation accuracy of the target recommendation users can be improved, and the method is helpful for helping customers to select satisfied articles. Meanwhile, based on the recommendation processing method provided by the application, the benefit improvement of the electronic commerce platform can be promoted. For ease of understanding, the present application uses actual data to verify the model effect. The data set extracted in the application is derived from real-world internet insurance data sales data, and comprises about 560 ten thousand pieces of data, and mainly comprises the following four data of recommended task scenes:

Special offer (scenario 1): for a recommended scenario for a user with a certain awareness, the number of samples is 390W.

Payment completion page (scenario 2): for the recommended scenario where the product purchasing user has completed, the sample size is 18.3W.

Policy list page (scenario 3): the user's recommended scene has not been purchased yet, sample number 41W.

Child product page (scenario 4): the product recommendation scenario prepared for children/children, sample number 115W.

The application takes a multitask learning model (selective adaptive multi-task network, SAMN) as a recommended value ordering model, takes an LR model as a base line, and respectively performs offline verification of a reordering algorithm on the data of the four recommended task scenes. Referring to fig. 4, fig. 4 is a schematic diagram of an offline verification result of a reordering algorithm according to an embodiment of the present application. As can be seen from the verification results in fig. 4, the reordering algorithm we designed achieves revenue improvement of 11.3%,6.1%,7.8% and 3% in four scenarios, respectively. The improvement was more pronounced than baseline, 14.7%,8.5%,10.5% and 4.5%, respectively.

Meanwhile, to verify the effectiveness of our reordering algorithm, we have deployed this invention in the real world and conducted 7 days of AB testing. Referring to fig. 5, fig. 5 is a schematic diagram illustrating a verification result of the AB test provided in the embodiment of the present application. As can be seen from the online verification results in fig. 5, the reordering algorithm (on same basis) in the embodiment of the present application achieves 16.2%,7.2%,7.15% and 2.7% revenue improvement in four scenarios, respectively, compared to baseline.

Referring to fig. 6, fig. 6 is a schematic structural diagram of a recommendation processing apparatus according to an embodiment of the present application. The recommendation processing device provided by the embodiment of the application comprises:

the fusion characteristic data determining module 31 is configured to obtain fusion characteristic data of each element to be recommended in the element set to be recommended, where the fusion characteristic data includes user characteristic data of a target recommended user;

the predicted ranking value determining module 32 is configured to rank and predict each element to be recommended in the element set to be recommended according to the fusion feature data of each element to be recommended, so as to obtain a predicted ranking value of each element to be recommended;

a reordering factor obtaining module 33, configured to perform a ranking optimization process based on historical feature data of each element to be recommended in the element set to be recommended, to obtain a reordering factor sequence, where the reordering factor sequence includes reordering factors corresponding to each element to be recommended, and the historical feature data includes user feature data of a sample user;

the recommendation ranking value determining module 34 is configured to determine a recommendation ranking value of each element to be recommended according to the predicted ranking value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine a recommendation ranking order of each element to be recommended in the element set to be recommended according to the recommendation ranking values.

Referring to fig. 7, fig. 7 is another schematic structural diagram of the recommendation processing apparatus according to the embodiment of the present application. Wherein:

in some possible implementations, the prediction ranking value determination module 32 is configured to:

inputting the fusion characteristic data of each element to be recommended into a ranking value prediction model, and performing ranking prediction on each element to be recommended in the element set to be recommended through the ranking value prediction model to obtain a predicted ranking value of each element to be recommended;

the reorder factor obtaining module 33 includes:

the historical ranking value set obtaining unit 331 is configured to input the historical feature data of each element to be recommended in the element set to be recommended to the ranking value prediction model to obtain a historical ranking value set corresponding to each element to be recommended, where the historical ranking value set corresponding to any element to be recommended includes: n historical ranking values of any element to be recommended to N sample users are obtained, wherein N is an integer greater than 0;

the optimization processing unit 332 is configured to perform a reordering optimization process based on the historical ordering value set corresponding to each element to be recommended, so as to obtain a reordering factor sequence.

In some possible implementations, the optimization processing unit 332 includes:

An equivalent ranking value determining subunit 3321, configured to determine an equivalent ranking value H (i) of each element to be recommended according to the historical ranking value set of each element to be recommended;

a first processing subunit 3322, configured to obtain a total exposure number I (I) of each element to be recommended to N sample users, and obtain a value P (I) of each element to be recommended;

a total conversion value determining subunit 3323, configured to determine a total conversion value Rev corresponding to the set of elements to be recommended according to the equivalent ranking value H (I) of each element to be recommended, the total exposure number I (I) of each element to be recommended to N sample users, and the price P (I) of each element to be recommended;

a reorder factor sequence determination subunit 3324 for determining a reorder factor sequence based on the total conversion value Rev.

In some possible embodiments, the total conversion value Rev satisfies:

wherein I (I) represents the total exposure number of each element I to be recommended to N sample users, H (I) represents the equivalent ranking value of each element I to be recommended, and P (I) represents the price of each element I to be recommended.

In some possible embodiments, the first processing subunit 3322 is specifically configured to:

Acquiring the exposure number I of each element to be recommended to each sample user u _u (i)；

Determining the total exposure number I (I) of each element to be recommended to all sample users by the sum of the exposure numbers of each element to be recommended to each sample user;

when the sorting sequence number of any element to be recommended corresponding to any sample user is smaller than or equal to M, the exposure number of any sample user to be recommended is equal to 1, and when the sorting sequence number of any element to be recommended corresponding to any sample user is larger than M, the exposure number of any sample user to be recommended is equal to 0, wherein M is determined according to the maximum exposure number.

In some possible embodiments, the equivalent rank value determining subunit 3321 is specifically configured to:

acquiring the association information set of each element to be recommended, wherein the association information set of any element to be recommended comprises: association information between any element to be recommended and each sample user of the N sample users;

acquiring an initial ranking factor sequence, and determining an equivalent ranking value H (i) of each element to be recommended according to the initial ranking factor of each element to be recommended in the initial ranking factor sequence, a historical ranking value set corresponding to each element to be recommended and an associated information set of each element to be recommended.

In some possible embodiments, the association information between the any element to be recommended and each sample user of the N sample users includes: the exposure number of any element to be recommended to each sample user;

the equivalent ranking value H (i) of each element to be recommended satisfies the following conditions:

wherein, gamma (i) represents the initial reordering factor corresponding to each element i to be recommended, CR _u (i) Representing the historical ranking value of each element to be recommended I for each sample user, I _u (i) Representing the exposure number of each element to be recommended to each sample user, ρ (rank (γ (i)) CR _u (i) -a) representing a conversion deviation coefficient for each of said elements i to be recommended;

the conversion deviation coefficient of each element to be recommended in each sample user comprises the following steps: according to the sequence number of each element to be recommended in the element set to be recommended, determining a coefficient from at least one candidate conversion deviation coefficient corresponding to each element to be recommended; the ranking number of any element to be recommended is determined according to the product of the initial reordering factor corresponding to any element to be recommended and the historical ranking value.

In some possible embodiments, the reorder factor sequence determination subunit 3324 is specifically configured to:

Detecting whether the sum of average income change values of all elements to be recommended meets a stop condition;

if yes, determining a new reordering factor sequence based on an updating rule of the reordering factor sequence and the initial reordering factor sequence, and performing ordering optimization processing based on the new reordering factor sequence.

In some possible implementations, the reorder factor update rule is determined from a dot-division result of the first reorder factor sequence and the second reorder factor sequence, a learning rate α, and an average revenue change vector Δr (t);

the first reordering factor sequence refers to: a reordering factor sequence obtained by the last iteration corresponding to the current initial reordering factor sequence;

the second reordering factor sequence refers to: and the reordering factor sequence obtained by the previous iteration corresponds to the reordering factor sequence obtained by the previous iteration.

In some possible embodiments, the recommendation-ranking-value determining module 34 is specifically configured to:

multiplying the predicted sorting value of each element to be recommended by the reordering factor to obtain the recommended sorting value of each element to be recommended.

Based on the same inventive concept, the principle and beneficial effects of the solution to the problem of the recommended processing device provided in the embodiments of the present application are similar to those of the message processing method in the embodiments of the present application, and may refer to the principle and beneficial effects of implementation of the method, and the relationships between the steps executed by each relevant module may refer to the descriptions of the relevant content in the foregoing embodiments, which are not repeated herein for brevity.

Referring to fig. 8, fig. 8 is a schematic structural diagram of a terminal device according to an embodiment of the present application. As shown in fig. 8, the terminal device in the present embodiment may include: one or more processors 401 and a memory 402. The processor 401 and the memory 402 are connected via a bus 403. The memory 402 is used for storing a computer program comprising program instructions, and the processor 401 is used for executing the program instructions stored in the memory 402 for performing the following operations:

acquiring fusion characteristic data of each element to be recommended in an element set to be recommended, wherein the fusion characteristic data comprises user characteristic data of a target recommended user;

sequencing and predicting each element to be recommended in the element to be recommended set according to the fusion characteristic data of each element to be recommended, and obtaining a predicted sequencing value of each element to be recommended;

And determining a recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value.

In some possible embodiments, the processor 401 is configured to:

the above processor 401 is further configured to:

inputting the historical characteristic data of each element to be recommended in the element set to be recommended into the ranking value prediction model to obtain a historical ranking value set corresponding to each element to be recommended, wherein the historical ranking value set corresponding to any element to be recommended comprises: n historical ranking values of any element to be recommended to N sample users are obtained, wherein N is an integer greater than 0;

and (3) carrying out reordering optimization processing based on the historical ordering value sets corresponding to each element to be recommended to obtain a reordering factor sequence.

In some possible embodiments, the processor 401 is configured to:

determining an equivalent ranking value H (i) of each element to be recommended according to the historical ranking value set of each element to be recommended;

acquiring the total exposure number I (I) of each element to be recommended to N sample users, and acquiring the value P (I) of each element to be recommended;

determining a total conversion value Rev corresponding to the element set to be recommended according to an equivalent ranking value H (I) of each element to be recommended, total exposure numbers I (I) of each element to be recommended to N sample users, and prices P (I) of each element to be recommended;

a reorder factor sequence is determined based on the total conversion value Re v.

In some possible embodiments, the total conversion value Rev satisfies:

In some possible embodiments, the processor 401 is configured to:

In some possible embodiments, the above processor 401 is further configured to:

It should be appreciated that in some possible embodiments, the processor 401 described above may be a central processing unit (central processing unit, CPU), which may also be other general purpose processors, digital signal processors (digital signal processor, DSP), application specific integrated circuits (application specific integrated circuit, ASIC), off-the-shelf programmable gate arrays (field programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The memory 402 may include read only memory and random access memory and provides instructions and data to the processor 401. A portion of memory 402 may also include non-volatile random access memory. For example, the memory 402 may also store information of device type.

In a specific implementation, the terminal device may execute, through each function module built in the terminal device, an implementation manner provided by each step in fig. 2 to 3, and specifically, the implementation manner provided by each step may be referred to, which is not described herein again.

Based on the same inventive concept, the principle and beneficial effects of the solution to the problem of the intelligent device provided in the embodiments of the present application are similar to those of the message processing method in the embodiments of the present application, and may refer to the principle and beneficial effects of implementation of the method, and the relationship between each step executed by the processor 401 may refer to the description of the related content in the foregoing embodiments, which is not repeated herein for brevity.

The embodiment of the present application further provides a computer readable storage medium, where a computer program is stored, where the computer program includes program instructions, and when the program instructions are executed by a processor, implement the recommended processing method provided by each step in fig. 2 to 3, and specifically, reference may be made to the implementation manners provided by each step, which are not described herein again.

The computer readable storage medium may be the recommendation processing apparatus provided in any one of the foregoing embodiments or an internal storage unit of the terminal device, for example, a hard disk or a memory of an electronic device. The computer readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash card (flash card) or the like, which are provided on the electronic device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of the electronic device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

The present application also provides a computer program product containing instructions, which when run on a computer, cause the computer to perform the recommended processing method described in the above method embodiment.

Embodiments of the present application also provide a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the computer device performs the method of the recommended processing described above.

The steps in the method of the embodiment of the application can be sequentially adjusted, combined and deleted according to actual needs.

The modules in the device of the embodiment of the application can be combined, divided and deleted according to actual needs.

The terms "first," "second," "third," "fourth," and the like in the claims and in the description and drawings of the present application, are used for distinguishing between different objects and not for describing a particular sequential order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed steps or elements but may include other steps or elements not listed or inherent to such process, method, article, or apparatus.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments. The term "and/or" as used in this specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations. Those of ordinary skill in the art will appreciate that the elements and algorithm steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, in computer software, or in a combination of the two, and that the elements and steps of the examples have been generally described in terms of function in the foregoing description to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

The methods and related devices provided in the embodiments of the present application are described with reference to the method flowcharts and/or structure diagrams provided in the embodiments of the present application, and each flowchart and/or block of the method flowcharts and/or structure diagrams may be implemented by computer program instructions, and combinations of flowcharts and/or blocks in the flowchart and/or block diagrams. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or structural diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or structures.

Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be implemented by a program to instruct related hardware, the program may be stored in a computer readable storage medium, and the readable storage medium may include: flash disk, read-Only Memory (ROM), random-access Memory (Random Access Memory, RAM), magnetic or optical disk, and the like.

The above disclosure is illustrative of a preferred embodiment of the present application and, of course, should not be taken as limiting the scope of the invention, and those skilled in the art will recognize that all or part of the above embodiments can be practiced with modification within the spirit and scope of the appended claims.

Claims

1. A recommendation processing method, the method comprising:

acquiring fusion characteristic data of each element to be recommended in an element set to be recommended, wherein the fusion characteristic data of one element to be recommended comprises user characteristic data of a target recommended user and element characteristic data of the one element to be recommended;

performing sequencing optimization processing based on historical feature data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended, and one piece of historical feature data comprises user feature data of one sample user and element feature data of the one element to be recommended;

determining a recommendation ordering value of each element to be recommended according to the prediction ordering value of each element to be recommended and the reordering factor of each element to be recommended, so as to determine the recommendation ordering sequence of each element to be recommended in the element set to be recommended according to the recommendation ordering value;

the sorting optimization processing is performed based on the historical feature data of each element to be recommended in the element set to be recommended, so as to obtain a reordering factor sequence, which comprises the following steps: inputting the historical characteristic data of each element to be recommended in the element set to be recommended into a ranking value prediction model to obtain a historical ranking value set corresponding to each element to be recommended, wherein the historical ranking value set corresponding to any element to be recommended comprises: n historical ranking values of any element to be recommended to N sample users are obtained, wherein N is an integer greater than 0; performing reordering optimization processing based on the historical ordering value sets corresponding to the elements to be recommended to obtain a reordering factor sequence;

The reordering optimization processing is performed based on the historical ordering value set corresponding to each element to be recommended, so as to obtain a reordering factor sequence, which comprises the following steps: determining the equivalent ranking value of each element to be recommended according to the historical ranking value set of each element to be recommendedThe method comprises the steps of carrying out a first treatment on the surface of the Acquiring the total exposure number of each element to be recommended to N sample users>And obtaining the value +/of each of the elements to be recommended>The method comprises the steps of carrying out a first treatment on the surface of the According to the equivalent ranking value of each element to be recommended +.>Total exposure number of each element to be recommended to N sample users +.>And the price of the individual element to be recommended +.>Determining the total conversion value corresponding to the set of elements to be recommended>The method comprises the steps of carrying out a first treatment on the surface of the Based on the total transformation value->Determining a reordering factor sequence; wherein the equivalent ranking value of the element i to be recommended +.>Determining, based on the historical ranking value set of the element to be recommended i, association information between any element to be recommended and each sample user of the N sample users includes: the exposure number of any element to be recommended to each sample user; the equivalent ranking value +.>The method meets the following conditions:

wherein,representing each of the elements to be recommendediCorresponding initial reordering factor,/- >Representing each of the elements to be recommendediFor each sample useruHistorical ranking value of->Representing the exposure number of each element to be recommended to each sample user,/for each sample user>Representing each of the elements to be recommendediIs a conversion deviation coefficient of (2); the conversion deviation coefficient of each element to be recommended in each sample user comprises the following steps: according to the sequence number of each element to be recommended in the element set to be recommended, determining a coefficient from at least one candidate conversion deviation coefficient corresponding to each element to be recommended; the sorting sequence number of any element to be recommended is determined according to the product of the initial reordering factor corresponding to any element to be recommended and the historical sorting value;

the determining the recommended ranking value of each element to be recommended according to the predicted ranking value of each element to be recommended and the reordering factor of each element to be recommended comprises the following steps: multiplying the predicted sorting value of each element to be recommended by the reordering factor to obtain the recommended sorting value of each element to be recommended.

2. The method of claim 1, wherein the predicting the ranking of each element to be recommended in the set of elements to be recommended according to the fused feature data of each element to be recommended to obtain the predicted ranking value of each element to be recommended comprises:

And inputting the fusion characteristic data of each element to be recommended into a ranking value prediction model, and performing ranking prediction on each element to be recommended in the element set to be recommended through the ranking value prediction model to obtain a predicted ranking value of each element to be recommended.

3. The method of claim 1, wherein the total conversion valueThe method meets the following conditions:

wherein the saidRepresenting the total exposure number of each element i to be recommended to N sample users,/for each element i to be recommended>Representing an equivalent ranking value of each of said elements i to be recommended,/->Representing the price of each element i to be recommended.

4. The method of claim 1, wherein the obtaining of the total exposure of each of the elements to be recommended to N sample usersComprising:

acquiring the exposure number of each element to be recommended to each sample user u；

Determining the total exposure number of each element to be recommended to all sample users by summing the exposure numbers of each element to be recommended to each sample user；

5. The method of any one of claims 1-4, wherein the determining an equivalent ranking value for each of the elements to be recommended is based on a set of historical ranking values for each of the elements to be recommendedComprising:

acquiring an initial ranking factor sequence, and according to the initial ranking factor of each element to be recommended in the initial ranking factor sequence and each element to be recommendedA historical ranking value set corresponding to the recommended elements and an associated information set of each element to be recommended determine an equivalent ranking value of each element to be recommended。

6. The method of claim 5, wherein the step of determining the total conversion value is based onDetermining a reordering factor sequence, further comprising:

7. The method of claim 6, wherein the reorder factor update rule is based on a division result of a first reorder factor sequence and a second reorder factor sequence, a learning rateAverage revenue change vector->Determining;

8. A recommendation processing apparatus, the apparatus comprising:

the fusion characteristic data determining module is used for acquiring fusion characteristic data of each element to be recommended in the element set to be recommended, wherein the fusion characteristic data of one element to be recommended comprises user characteristic data of a target recommended user and element characteristic data of the one element to be recommended;

the prediction ranking value determining module is used for performing ranking prediction on each element to be recommended in the element set to be recommended according to the fusion characteristic data of each element to be recommended to obtain a prediction ranking value of each element to be recommended;

The reordering factor acquisition module is used for carrying out ordering optimization processing based on the historical characteristic data of each element to be recommended in the element set to be recommended to obtain a reordering factor sequence, wherein the reordering factor sequence comprises reordering factors corresponding to each element to be recommended, and one piece of historical characteristic data comprises user characteristic data of one sample user and element characteristic data of the one element to be recommended;

the recommendation ranking value determining module is used for determining a recommendation ranking value of each element to be recommended according to the prediction ranking value of each element to be recommended and the reordering factor of each element to be recommended so as to determine the recommendation ranking sequence of each element to be recommended in the element set to be recommended according to the recommendation ranking value;

wherein, the reordering factor acquisition module includes: the history ranking value set obtaining unit is configured to input history feature data of each element to be recommended in the element set to be recommended to a ranking value prediction model to obtain a history ranking value set corresponding to each element to be recommended, where the history ranking value set corresponding to any element to be recommended includes: n historical ranking values of any element to be recommended to N sample users are obtained, wherein N is an integer greater than 0; the optimizing processing unit is used for carrying out reordering optimizing processing based on the historical ordering value sets corresponding to the elements to be recommended to obtain a reordering factor sequence;

Wherein the optimization processing unit includes: an equivalent ranking value determining subunit for determining, based on eachThe historical ranking value set of each element to be recommended determines the equivalent ranking value of each element to be recommendedThe method comprises the steps of carrying out a first treatment on the surface of the A first processing subunit for acquiring the total exposure number of each element to be recommended to N sample users>And obtaining the value +/of each of the elements to be recommended>The method comprises the steps of carrying out a first treatment on the surface of the A total conversion value determination subunit for determining +.A total conversion value of each element to be recommended based on>Total exposure number of each element to be recommended to N sample users +.>And the price of the individual element to be recommended +.>Determining the total conversion value corresponding to the set of elements to be recommended>The method comprises the steps of carrying out a first treatment on the surface of the A reordering factor sequence determining subunit for +.>Determining a reordering factor sequence; wherein the equivalent ranking value of the element i to be recommended +.>Determining, based on the historical ranking value set of the element to be recommended i, association information between any element to be recommended and each sample user of the N sample users includes: for each sample, any element to be recommendedThe number of exposures of the user; the equivalent ranking value +.>The method meets the following conditions:

wherein, Representing each of the elements to be recommendediCorresponding initial reordering factor,/->Representing each of the elements to be recommendediFor each sample useruHistorical ranking value of->Representing the exposure number of each element to be recommended to each sample user,/for each sample user>Representing each of the elements to be recommendediIs a conversion deviation coefficient of (2); the conversion deviation coefficient of each element to be recommended in each sample user comprises the following steps: according to the sequence number of each element to be recommended in the element set to be recommended, determining a coefficient from at least one candidate conversion deviation coefficient corresponding to each element to be recommended; the sorting sequence number of any element to be recommended is determined according to the product of the initial reordering factor corresponding to any element to be recommended and the historical sorting value;

the recommendation ordering value determining module is specifically configured to: multiplying the predicted sorting value of each element to be recommended by the reordering factor to obtain the recommended sorting value of each element to be recommended.

9. A terminal device comprising a processor and a memory, said processor and said memory being interconnected;

the memory is for storing a computer program comprising program instructions, the processor being configured to invoke the program instructions to perform the method of any of claims 1-7.

10. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the method of any of claims 1-7.